Reference table
LLM API Pricing Comparison 2026: Price History per 1M Tokens Since GPT-4
As of 2026-11-03, OpenAI's cheapest frontier-tier model lists at $1.75 per 1M input tokens, against $30 for GPT-4 in 2023 (dated). Six vendors.
As of 2026-11-03, the cheapest frontier-tier input price per vendor ranges from $0.70 per 1M tokens (Mistral AI's Mistral Large 4) to $4 per 1M tokens (Anthropic's Claude Opus 5.5). S33 S15
As of 2026-11-03, the cheapest frontier-tier LLM API input price per vendor ranges from $0.70 S33 per 1M tokens (Mistral AI’s Mistral Large 4) to $4 S15 per 1M tokens (Anthropic’s Claude Opus 5.5). OpenAI listed GPT-4 (8K context) at $30 S02 per 1M input tokens on 2023-03-15 (dated). As of 2026-11-03 its cheapest frontier-tier model, gpt-5.2, lists at $1.75 S01, which is 94.2% derived lower than the dated GPT-4 price. The largest same-name cut in the data is OpenAI o3, down 80.0% derived from $10.00 S09 to $2.00 S10 per 1M input tokens between 2025-04-19 and 2025-07-02. Prices come from the pricing pages of OpenAI, Anthropic, Google, Mistral AI, DeepSeek and xAI and from archived snapshots of them.
Open the LLM API price explorer
The explorer draws every publishable list price as a step line per model family, with filters and a sortable table. It runs in your browser with no tracking. To embed it, paste this code. Keep the attribution line under the iframe. The last line fits the frame to the explorer; if your site strips scripts, leave it out and the frame scrolls inside.
<iframe src="https://sysprobs.com/data/tools/llm-api-price-history/embed/" width="100%" height="820" loading="lazy" style="border:0;max-width:100%" title="LLM API Price History Tracker: interactive chart of LLM API prices per 1M tokens"></iframe>
<p>Chart: <a href="https://sysprobs.com/data/llm-api-price-history/">LLM API Price History Tracker</a> by Sysprobs Data. Vendor list prices, data as of 2026-11-03.</p>
<script>window.addEventListener("message",function(e){if(e.origin!=="https://sysprobs.com"||!e.data||!e.data.sysprobsEmbedHeight)return;var f=document.getElementsByTagName("iframe");for(var i=0;i<f.length;i++){if(f[i].contentWindow===e.source)f[i].style.height=e.data.sysprobsEmbedHeight+"px"}});</script>
LLM API pricing comparison as of 2026-11-03
Each row is the model with the lowest standard input price per tier on 2026-11-03, under the tier definition below. Models closed to new customers do not count. Prices are US dollars per 1M tokens, standard rate, short prompts.
| Vendor | Cheapest frontier-tier model | Input | Output | Cheapest small-tier model | Input | Output |
|---|---|---|---|---|---|---|
| OpenAI | gpt-5.2 | $1.75 S01 | $14.00 S01 | gpt-6-luna | $0.10 S01 | $0.50 S01 |
| Anthropic | Claude Opus 5.5 | $4 S15 | $20 S15 | Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 S15 | $0.50 S15 |
| Gemini 3.1 Pro Preview and Gemini 4 Argon (tied on input) | $2.00 S25 | $12.00 S25 and $10.00 S25 | Gemini 3.1 Flash-Lite | $0.25 S25 | $1.50 S25 | |
| Mistral AI | Mistral Large 4 | $0.70 S33 | $2.10 S33 | Mistral Small 4 | $0.15 S33 | $0.60 S33 |
| DeepSeek | deepseek-v4-pro (peak rate) | $1.32 S38 | $3.96 S38 | deepseek-flash (peak rate) | $0.30 S38 | $1.20 S38 |
| xAI | Grok 4.7 (availability unknown) | $2.00 S43 | $6.00 S43 | no small model displayed | n/a | n/a |
Google’s deprecations page S56 limits access to the 2.5 models to users who have actively used them in the past. On 2026-11-03 it also lists gemini-2.5-pro and gemini-2.5-flash as deprecated, with a shutdown date of 2026-12-01, and says gemini-2.5-flash-lite is not deprecated. Gemini 2.5 Pro and Gemini 2.5 Flash-Lite are therefore closed to new customers and are not in the table. Their list prices on 2026-11-03 are $1.25 S25 input and $10.00 S25 output for 2.5 Pro, and $0.10 S25 and $0.40 S25 for 2.5 Flash-Lite. xAI’s page records no deprecations or access limits, so Grok 4.7’s availability to new customers is unknown and it still counts.
Claude Haiku 5.5 is first listed in our 2026-10-20 snapshot and enters Anthropic’s small tier from that snapshot. Anthropic prices it by prompt length: $0.10 S15 input and $0.50 S15 output per 1M tokens up to 100,000 tokens, and $0.50 S15 and $2.50 S15 above that. Claude Haiku 4.5, at $1 S15 and $5 S15, was in the current model table on 2026-10-07. On 2026-10-20 and again on 2026-11-03 Anthropic’s pricing page lists it under Legacy models at an unchanged price, so it no longer counts.
Gemini 4 Argon is absent from Google’s pricing page on 2026-10-07. On 2026-10-20 the page listed it at $2.00 S25 input and $10.00 S25 output per 1M tokens through 2026-11-30, and it lists the same prices on 2026-11-03. Google’s 2026-11-03 copy of the page shows no update date. A scheduled change takes the price to $4.00 S25 and $20.00 S25 starting 2026-12-01; that price is not a series value before then. Google’s pricing page S25 calls Argon its flagship reasoning model. It is not Pro branded, and it counts as Google’s frontier model by an editorial choice of 2026-10-22. On 2026-10-20 and 2026-11-03 it ties with Gemini 3.1 Pro Preview at $2.00 S25 input, so the table names both.
Lower list prices sit on the same pages for other models the definition leaves out. OpenAI lists gpt-5 and gpt-5.1 at $1.25 S01 input. OpenAI’s deprecations page S14 lists gpt-5 (gpt-5-2025-08-07) for removal on 2026-12-11 and gpt-5.1 (gpt-5.1-2025-11-13) for removal on 2027-04-01, in entries dated 2026-06-11 and 2026-10-01. gpt-5.3-codex lists at $1.75 S01 but is a specialised model. gpt-6.1-terra, new on 2026-11-03, lists at $1.00 S01 input and $5.00 S01 output, but terra is OpenAI’s mid tier. Claude Opus 4.1 lists at $15 S15 input and $75 S15 output, in Anthropic’s legacy table.
Key findings
- As of 2026-11-03, OpenAI’s cheapest frontier-tier model is gpt-5.2 at $1.75 S01 per 1M input tokens and $14.00 S01 output.
- GPT-4 (8K context) listed at $30 S02 input and $60 S02 output on 2023-03-15 (dated). Against gpt-5.2, that is 94.2% derived lower on input and 76.7% derived on output, both dated comparisons.
- As of 2026-11-03, Google’s cheapest frontier-tier input price open to new customers is $2.00 S25, a tie between Gemini 3.1 Pro Preview and Gemini 4 Argon. Gemini 2.5 Pro lists at $1.25 S25 but is closed to new customers.
- OpenAI cut o3 by 80.0% derived between the snapshots of 2025-04-19 and 2025-07-02. Input fell from $10.00 S09 to $2.00 S10 per 1M tokens, output from $40.00 S09 to $8.00 S10.
- Google cut Gemini 1.5 Flash input by 78.6% derived (dated), from $0.35 S26 per 1M tokens on 2024-05-30 (dated) to $0.075 S27 on 2024-08-29 (dated).
- The lowest standard input price in the dataset is $0.0375 S28 per 1M tokens, for Gemini 1.5 Flash-8B on 2024-12-03.
- Anthropic’s cheapest frontier-tier price fell from $15 S17 input and $75 S17 output (Claude 3 Opus, 2024-03-04, dated) to $4 S15 and $20 S15 (Claude Opus 5.5, 2026-11-03).
- DeepSeek’s deepseek-chat list output price was $0.28 S39 on 2024-05-28 (dated), $1.10 S40 on 2025-01-24 and $1.68 S41 on 2025-09-10, then $0.42 S42 on 2026-03-01. A promotional $0.28 S40 applied until 2025-02-08.
- The gpt-4o alias halved in price without a name change: $5.00 S07 per 1M input tokens on 2024-08-31 (dated), $2.50 S08 on 2024-12-14.
- Google’s pricing page lists Gemini 3.6 Flash and Gemini 3.8 Flash at $0.75 S25 per 1M input tokens on 2026-11-03. A scheduled change takes the price to $1.50 S25 from 2027-01-01. Gemini 3.7 Flash, listed at $0.75 S25 on 2026-10-07, is not on that page on 2026-10-20 or 2026-11-03.
- OpenAI’s deprecations page S14 records the shutdown of o1-preview on 2025-07-28 and of o1-mini on 2025-10-27. It says access to gpt-4, gpt-4-turbo, gpt-3.5-turbo, gpt-4.1-nano, o1, o3-mini, o4-mini and other models was shut down on 2026-10-23.
- Mistral Large 4 lists at $0.70 S33 input and $2.10 S33 output on 2026-11-03, with no sale label. Mistral Large 3, $0.50 S33 and $1.50 S33 on 2026-10-20, is off the pricing page and deprecated from 2026-10-31.
How the tiers are defined
The charts and the comparison table follow one definition, quoted from the dataset methodology.
A model counts for a vendor’s frontier or small tier on a snapshot date when all of these hold:
- Its family is in the vendor’s tier set below.
- The row is the standard-tier, short-context, USD vendor list price, with both an input and an output price on that snapshot.
- The page does not place it in a legacy or older-models section and does not label it legacy or deprecated.
- The vendor’s deprecations page shows no deprecation announced on or before the snapshot date, and no shutdown date on or before it.
- It is not a specialised model: cyber, codex, chat-latest, search, research, realtime, audio, live, image, speech, robotics or computer-use variants are excluded.
- The price used is the list price in effect on that date: not a promotional price and not a scheduled future price.
- It is not closed to new customers on that date. Unknown availability does not exclude a model.
| Vendor | Frontier tier | Small tier | Neither |
|---|---|---|---|
| OpenAI | GPT-branded text models with no size suffix or with the sol or astra suffix: GPT-4, GPT-4 Turbo, GPT-4o, GPT-4.1, GPT-5, GPT-5.1, GPT-5.2, GPT-5.4, GPT-5.5, gpt-5.6-sol, gpt-6-sol, gpt-6.1-sol, gpt-6-astra | GPT-3.5 Turbo (4K and 16K) and the mini, nano and luna models | o-series reasoning models, pro variants, terra (mid tier), InstructGPT and base models, specialised models |
| Anthropic | Opus-branded models and Fable; Claude 2 and 2.1 before Opus existed | Claude Instant and Haiku models | Sonnet (mid tier); gated Mythos models |
| Pro text models from Gemini 1.5 on (Gemini 2.5 Pro counts until it is closed to new customers), and Gemini 4 Argon (editorial choice of 2026-10-22) | Flash, Flash-8B and Flash-Lite text models (Gemini 2.5 Flash and Flash-Lite count until closed to new customers) | Gemini 1.0 Pro; Live, native audio, image, speech, robotics and computer-use variants | |
| Mistral AI | Mistral Large of any generation, at the original price | Mistral Small of any generation | Medium (mid tier), Ministral edge models, the open 7B, NeMo, Mixtral and Pixtral models, Codestral, Devstral, Saba |
| DeepSeek | deepseek-reasoner and deepseek-v4-pro, peak rate | deepseek-chat and deepseek-flash, peak rate | none |
| xAI | Grok 3, Grok 4 and Grok 4.7 | grok-3-mini | grok-code-fast-1; models that appear only in the page’s model registry |
Legacy status comes from the page’s own section or label. Deprecation comes from the vendor’s deprecations page. Access limits come from a line saying only existing users can use the model. If several models share the lowest input price, the dataset keeps every one. On the input charts the hover text of that point names every tied model. The output charts plot the lowest output price among the tied models. The hover text names that model and gives the other model’s price, which is also in the CSV. Each quarter-end point uses the last snapshot in that quarter.
Frontier-tier LLM API prices over time
- Tier sets follow vendor naming. Legacy, deprecated, specialised and promotion-only models are excluded (see methodology).
- List prices only: promotional and off-peak prices are not series values. Gaps are quarters with no readable snapshot.
- A 2026-Q4 point is the 2026-11-03 snapshot (quarter to date); a 2026-Q3 point is the 2026-10-07 snapshot.
- Points of one vendor that would overlap are drawn side by side, up to 8 days from their true date; the hover text gives the true date.
- Model names for every point are in the hover text, the chart description and the CSV; each panel heading names the vendor's latest point.
- Hollow dashed rings are cheaper models closed to new customers. They are not part of the line, which shows through them.
- Derived rows (source_id DERIVED) in the CSV download name the snapshot key, the candidate models and the tier definition for every point.
- Each vendor has its own panel, in a grid on wide screens and one under another on phones; every panel has the same price scale and the same years.
OpenAI. GPT-4 Turbo (gpt-4-1106-preview) listed at $0.01 S04 per 1K input tokens on 2023-11-14 (dated), or $10 per 1M after conversion. gpt-4o listed at $5.00 S06 on 2024-05-31 (dated) and the August gpt-4o snapshot at $2.50 S07 (dated). GPT-4.1 listed at $2.00 S09 on 2025-04-19, GPT-5 at $1.250 S11 on 2025-08-31 and GPT-5.2 at $1.750 S12 on 2025-12-14. As of 2026-10-07, gpt-5.2 lists at $1.75 S01. gpt-6.1-sol lists at $2.00 S01 input and $10.00 S01 output. Against GPT-4 (8K context), gpt-6.1-sol is 93.3% derived lower on input and 83.3% derived on output, both dated comparisons.
Anthropic. Anthropic’s oldest captured price sheet, archived on 2023-09-13 (dated), lists Claude 2 at $11.02 S16 input and $32.68 S16 output per 1M tokens. Claude 2.0 and 2.1 listed at $8 S17 and $24 S17 on 2024-03-04 (dated). Anthropic’s launch post S60 dates the release of Claude 3 Opus to 2024-03-04. It listed at $15 S17 and $75 S17 that day (dated). Opus 4.1 still listed at $15 S23 and $75 S23 on 2025-10-04. That snapshot supplies both the 2025-Q3 and 2025-Q4 points, because no capture exists between 2025-10-04 and 2026-03-02. By 2026-03-02, Opus 4.6 listed at $5 S24 and $25 S24. As of 2026-10-07, Claude Opus 5.5 lists at $4 S15 and $20 S15.
Google. Gemini 1.5 Pro listed at $3.50 S26 input and $10.50 S26 output on 2024-05-30 (dated). Google announced a 64% S59 input cut and a 52% S59 output cut on 2024-09-24 (dated), effective 2024-10-01. The 2024-12-03 snapshot shows $1.25 S28 and $5.00 S28. From the 2025-12-31 snapshot to the 2026-05-31 snapshot, the Google point is Gemini 2.5 Pro at $1.25 S31 input and $10.00 S31 output. Google’s deprecations page S56 now limits access to Gemini 2.5 Pro to users who have used it before, and on 2026-11-03 it lists a shutdown date of 2026-12-01 for it. The 2026-Q3 point, from the 2026-10-07 snapshot, is therefore Gemini 3.1 Pro Preview at $2.00 S25 and $12.00 S25. The 2026-Q4 to date point, from the 2026-11-03 snapshot, is a tie on input at $2.00 S25. Gemini 3.1 Pro Preview lists $12.00 S25 output and Gemini 4 Argon $10.00 S25. Gemini 2.5 Pro at $1.25 S25 and $10.00 S25 is drawn as a hollow dashed ring outside the line, and its hover text says it is closed to new customers. Gemini 4 Argon, absent on 2026-10-07 and listed on 2026-10-20 and 2026-11-03, lists at $2.00 S25 input and $10.00 S25 output through 2026-11-30. From 2026-12-01 it lists at $4.00 S25 and $20.00 S25. It counts as Google’s frontier model by an editorial choice of 2026-10-22. Its scheduled $4.00 price is not a point on the chart.
Mistral AI. Mistral Large listed at $8 S34 input and $24 S34 output in the 2024-02-29 snapshot (dated). Mistral Large 2 listed at $2 S35 and $6 S35 in the 2024-09-19 snapshot (dated). As of 2026-10-07, Mistral Large 3 listed at $0.50 S33 and $1.50 S33, and Mistral Large 4 at an original price of $1.36 S33 and $4.18 S33. The 2026-Q3 point is Mistral Large 3, because the tier uses the original price while a sale price is shown. On 2026-11-03 Mistral’s pricing page no longer lists Mistral Large 3, and Mistral’s model table S33 shows it as deprecated from 2026-10-31. Mistral Large 4 lists at $0.70 S33 input and $2.10 S33 output, with no original or sale label, so it is the 2026-Q4 to date point. The point moves from $0.50 S33 to $0.70 S33 on input and from $1.50 S33 to $2.10 S33 on output because the model changes, not because Mistral repriced Mistral Large 3.
DeepSeek. The reasoning alias deepseek-reasoner listed at $0.55 S40 input and $2.19 S40 output on 2025-01-24. It was $0.28 S42 and $0.42 S42 on 2026-03-01. As of 2026-10-07, deepseek-v4-pro lists at $1.32 S38 input and $3.96 S38 output at the peak rate.
xAI. xAI listed grok-3 at $3.00 S45 input and $15.00 S45 output on 2025-09-11. As of 2026-10-07, Grok 4.7 lists at $2.00 S43 and $6.00 S43. It is the only language model with a displayed price on the live page. xAI’s page records no access limits, so its availability to new customers is unknown.
- Tier sets follow vendor naming. Legacy, deprecated, specialised and promotion-only models are excluded (see methodology).
- List prices only: promotional and off-peak prices are not series values. Gaps are quarters with no readable snapshot.
- A 2026-Q4 point is the 2026-11-03 snapshot (quarter to date); a 2026-Q3 point is the 2026-10-07 snapshot.
- Where two models tie on input, this chart plots the lower output price; the other model's price is in the hover text and the CSV.
- Points of one vendor that would overlap are drawn side by side, up to 49 days from their true date; the hover text gives the true date.
- Model names for every point are in the hover text, the chart description and the CSV; each panel heading names the vendor's latest point.
- Hollow dashed rings are cheaper models closed to new customers. They are not part of the line, which shows through them.
- Derived rows (source_id DERIVED) in the CSV download name the snapshot key, the candidate models and the tier definition for every point.
- Each vendor has its own panel, in a grid on wide screens and one under another on phones; every panel has the same price scale and the same years.
The chart picks the model with the lowest input price, so its output price can rise while input falls. OpenAI’s frontier-tier point listed $8.00 S09 output for GPT-4.1 on 2025-04-19 and $10.000 S11 for GPT-5 on 2025-08-31. It listed $14.000 S12 for GPT-5.2 on 2025-12-14 and $15.00 S13 for GPT-5.4 on 2026-06-01. As of 2026-10-07 it is $14.00 S01 for gpt-5.2.
Small-tier LLM API prices over time
- Tier sets follow vendor naming. Legacy, deprecated, specialised and promotion-only models are excluded (see methodology).
- List prices only: promotional and off-peak prices are not series values. Gaps are quarters with no readable snapshot.
- A 2026-Q4 point is the 2026-11-03 snapshot (quarter to date); a 2026-Q3 point is the 2026-10-07 snapshot.
- Points of one vendor that would overlap are drawn side by side, up to 8 days from their true date; the hover text gives the true date.
- Model names for every point are in the hover text, the chart description and the CSV; each panel heading names the vendor's latest point.
- Hollow dashed rings are cheaper models closed to new customers. They are not part of the line, which shows through them.
- Derived rows (source_id DERIVED) in the CSV download name the snapshot key, the candidate models and the tier definition for every point.
- Each vendor has its own panel, in a grid on wide screens and one under another on phones; every panel has the same price scale and the same years.
OpenAI. gpt-3.5-turbo had one combined price of $0.002 S02 per 1K tokens on 2023-03-15 (dated), or $2 per 1M after conversion. gpt-3.5-turbo-0125 listed at $0.50 S05 input on 2024-03-01 (dated), gpt-4o-mini at $0.150 S07 on 2024-08-31 (dated) and GPT-5 nano at $0.050 S11 on 2025-08-31. As of 2026-10-07, gpt-6-luna lists at $0.10 S01 input and $0.50 S01 output. The same page lists gpt-5-nano at $0.05 S01 input. It is excluded because OpenAI’s deprecations page S14 schedules its removal for 2026-12-11.
Anthropic. Claude Instant listed at $1.63 S16 input and $5.51 S16 output on 2023-09-13 (dated). Claude 3 Haiku listed at $0.25 S17 and $1.25 S17 on 2024-03-04 (dated). On the 2025-06-14 page, Claude Haiku 3 sits in the legacy section at $0.25 S22. The small-tier point therefore moves to Claude Haiku 3.5 at $0.80 S22 input and $4 S22 output. That step reflects the legacy label, not a price change. As of 2026-10-07, Claude Haiku 4.5 lists at $1 S15 and $5 S15. On 2026-10-20 and 2026-11-03 Anthropic’s pricing page lists it under Legacy models S15 at an unchanged price. Claude Haiku 5.5, first listed in our 2026-10-20 snapshot, is then the small-tier point, at $0.10 S15 input and $0.50 S15 output for prompts up to 100,000 tokens. Above that length it lists at $0.50 S15 and $2.50 S15.
Google. Gemini 1.5 Flash listed at $0.35 S26 input on 2024-05-30 (dated) and $0.075 S27 on 2024-08-29 (dated). Gemini 1.5 Flash-8B listed at $0.0375 S28 input and $0.15 S28 output on 2024-12-03. The 2025-Q4 point is Gemini 2.0 Flash-Lite at $0.075 S31 and $0.30 S31. Google’s deprecations page S56 now limits access to Gemini 2.5 Flash and 2.5 Flash-Lite to users who have used them before. On 2026-11-03 it lists Gemini 2.5 Flash for shutdown on 2026-12-01 and says Gemini 2.5 Flash-Lite is not deprecated. Gemini 2.5 Flash-Lite lists at $0.10 S25 input and $0.40 S25 output on 2026-10-07, 2026-10-20 and 2026-11-03. It is drawn as a hollow dashed ring outside the line, and its hover text says it is closed to new customers. The 2026-Q3 point is Gemini 3.1 Flash-Lite at $0.25 S25 and $1.50 S25. The 2026-Q4 to date point is the same. Gemini 3.5 Flash and Gemini 3.7 Flash, which were 2026-Q3 candidates, are not on Google’s pricing page on 2026-10-20 or 2026-11-03 S25. Google’s deprecations page S56 gives no shutdown date for either. It says the endpoints are not shutting down and route API requests to Gemini 3.6 Flash and Gemini 3.8 Flash respectively.
Mistral AI. Mistral Small fell from $2 S34 input on 2024-02-29 (dated) to $0.2 S35 for Small 24.09 on 2024-09-19 (dated) and $0.1 S36 for Small 3.1 on 2025-05-31. As of 2026-10-07, Mistral Small 4 lists at $0.15 S33 input and $0.60 S33 output. The Ministral edge models are outside the small tier.
DeepSeek and xAI. deepseek-chat listed at $0.14 S39 input on 2024-05-28 (dated) and $0.28 S42 on 2026-03-01. As of 2026-10-07, deepseek-flash lists at $0.30 S38 input and $1.20 S38 output at the peak rate. xAI listed grok-3-mini at $0.30 S45 input and $0.50 S45 output on 2025-09-11. The live xAI page displays no small model, so xAI has no 2026-Q3 small-tier point.
- Tier sets follow vendor naming. Legacy, deprecated, specialised and promotion-only models are excluded (see methodology).
- List prices only: promotional and off-peak prices are not series values. Gaps are quarters with no readable snapshot.
- A 2026-Q4 point is the 2026-11-03 snapshot (quarter to date); a 2026-Q3 point is the 2026-10-07 snapshot.
- Where two models tie on input, this chart plots the lower output price; the other model's price is in the hover text and the CSV.
- Points of one vendor that would overlap are drawn side by side, up to 8 days from their true date; the hover text gives the true date.
- Model names for every point are in the hover text, the chart description and the CSV; each panel heading names the vendor's latest point.
- Hollow dashed rings are cheaper models closed to new customers. They are not part of the line, which shows through them.
- Derived rows (source_id DERIVED) in the CSV download name the snapshot key, the candidate models and the tier definition for every point.
- Each vendor has its own panel, in a grid on wide screens and one under another on phones; every panel has the same price scale and the same years.
GPT-5 nano is missing from the 2025-12-14 capture. The 2025-Q4 OpenAI point is therefore GPT-5 mini at $2.000 S12 output, a gap in the capture rather than a price rise. Anthropic’s point fell from $5 S15 output for Claude Haiku 4.5 on 2026-10-07 to $0.50 S15 for Claude Haiku 5.5 on 2026-10-20 (prompts up to 100,000 tokens). That is a change of model: Claude Haiku 4.5 is listed under Legacy models on 2026-10-20 and 2026-11-03 at an unchanged price. The last point of each line is the 2026-Q4 quarter to date, so a later snapshot in the quarter replaces it.
The biggest LLM API price cuts
- xAI has no same-name repricing in the retrievable snapshots and is not shown.
- The Mistral Large cut was already in force on 2024-09-19 under the name Mistral Large 2.
Each bar is a vendor’s largest cut for one model name between consecutive snapshots. The cut is (old price minus new price) divided by old price, times 100.
- OpenAI o3: 80.0% derived, from $10.00 S09 to $2.00 S10 input, 2025-04-19 to 2025-07-02.
- Google Gemini 1.5 Flash: 78.6% derived (dated), from $0.35 S26 to $0.075 S27 input, 2024-05-30 to 2024-08-29. Output fell 71.4% derived (dated), from $1.05 S26 to $0.30 S27.
- Mistral AI Mistral Large: 75.0% derived (dated), from $8 S34 to $2 S36 input (2024-02-29 to 2025-05-31). Mistral Large 2 already lists at $2 S35 (dated) on the 2024-09-19 page, and the 2025-05-31 label does not state a generation.
- DeepSeek deepseek-chat: 75.0% derived on output, from $1.68 S41 to $0.42 S42, 2025-09-10 to 2026-03-01.
- Anthropic Claude Instant: 56.4% derived (dated) on output, from $5.51 S16 to $2.40 S17, 2023-09-13 to 2024-03-04.
xAI has no same-name repricing in its two retrievable snapshots. Some cuts were announced first: OpenAI’s DevDay post of 2023-11-06 (dated) put GPT-4 Turbo input at $0.01 S57 per 1K tokens.
LLM API price increases
Same model name. DeepSeek’s deepseek-chat list input price rose from $0.14 S39 on 2024-05-28 (dated) to $0.27 S40 on 2025-01-24 and $0.56 S41 on 2025-09-10. On 2025-09-10 one price covered both DeepSeek aliases. Mistral’s Codestral listed at $0.2 S35 input on 2024-09-19 (dated) and $0.3 S36 on 2025-05-31.
Preview to general availability. Gemini 2.5 Flash Preview input was $0.15 S29 on 2025-04-30. Gemini 2.5 Flash, a different model name, listed $0.30 S30 at general availability on 2025-07-31. The preview priced output two ways: $0.60 S29 without thinking and $3.50 S29 with thinking, per 1M tokens. The general availability page shows one output price, $2.50 S30, which includes thinking tokens. Like for like, thinking output fell 28.6% derived, computed as (3.50 minus 2.50) divided by 3.50. The non-thinking preview price has no counterpart at general availability, so it is not compared.
Newer models at higher prices. These are successor models, not repricings of one model. GPT-5 listed at $1.250 S11 input and $10.000 S11 output on 2025-08-31. GPT-5.2 listed at $1.750 S12 and $14.000 S12 on 2025-12-14, and GPT-5.5 at $5.00 S13 and $30.00 S13 on 2026-06-01. GPT-5 mini listed at $0.250 S11 input on 2025-08-31 and GPT-5.4 mini at $0.75 S13 on 2026-06-01.
Anthropic Haiku. Claude 3.5 Haiku listed at $1.00 S20 input and $5.00 S20 output on 2024-11-20. The same page listed Claude 3 Haiku at $0.25 S20 and $1.25 S20. Anthropic revised 3.5 Haiku to $0.80 S58 and $4 S58 on 2024-12-03.
Promotions and scheduled price changes
The charts, the comparison table and the key findings use the list price in effect on the date shown. A promotional, sale or time-of-day price is kept as a separate row in the dataset and noted here with its window. A scheduled future price is kept as a separate row with its effective date and is never shown before that date.
- Google Gemini 3.6 and 3.8 Flash. The price in effect on 2026-11-03 is $0.75 S25 input and $3.75 S25 output, which the page shows through 2026-12-31. A scheduled change takes the price to $1.50 S25 and $7.50 S25 starting 2027-01-01. Gemini 3.7 Flash showed the same prices and schedule on 2026-10-07 and is not on the page on 2026-10-20 or 2026-11-03.
- Google Gemini 4 Argon. The price in effect on 2026-11-03 is $2.00 S25 input and $10.00 S25 output, which the page shows through 2026-11-30. A scheduled change takes the price to $4.00 S25 and $20.00 S25 starting 2026-12-01.
- DeepSeek deepseek-chat, 2025-01-24. The list price was $0.27 S40 input and $1.10 S40 output, shown struck through. A promotional $0.14 S40 and $0.28 S40 applied until 16:00 UTC on 2025-02-08. deepseek-reasoner was not discounted.
- DeepSeek off-peak rates. As of 2026-10-07, deepseek-flash input is $0.30 S38 at peak and $0.15 S38 off-peak. Off-peak rates are half the peak rate under a rule in force from 2026-08-16, recorded in DeepSeek’s change log at api-docs.deepseek.com/updates.
- Mistral Large 4. The original price was $1.36 S33 input and $4.18 S33 output on 2026-10-07 and 2026-10-20. The sale price was $0.68 S33 and $2.09 S33, with no end date shown. On 2026-11-03 the page shows one price, $0.70 S33 input and $2.10 S33 output, and no sale label.
- OpenAI gpt-5.6-sol. The page shows only a promotional price of $4.00 S01 input and $20.00 S01 output, available at least through 2026-11-21. With no list price shown, gpt-5.6-sol has no point in the charts.
- OpenAI gpt-6-astra-law. The page shows only a promotional price of $10.00 S01 input and $50.00 S01 output for this legal configuration of gpt-6-astra, available at least through 2026-12-31. It has no point in the charts.
Model lifespans and retirements
- Month-only release dates start on the 1st. Families without a release date in the ledger are not drawn.
- Solid bars end at a retirement or shutdown date from the vendor's deprecation page (Google gives earliest possible dates).
- Lighter bars have no retirement date in the sources and stop at the snapshot date.
- A dashed hollow end marker and the words closed to new customers mean the vendor's page says only existing users can use the model.
- Start dates have four bases, named in each bar's tooltip. Two are a launch post and a model ID date suffix (a build date that can precede public availability). The others are the release date column of the vendor's deprecations page and a version number that encodes a month.
- Claude 3 Opus and Sonnet start on their launch post date. Claude 3 Haiku starts on its model ID date, because no primary source in the ledger dates its availability.
- Dates and prices: the explorer's table. The lifespan of each family with a retirement date is also a derived row in the CSV download.
Each bar runs from a model family’s release date to its retirement or shutdown date. The start date comes from a launch post, a model ID date suffix or a deprecations page release date, and each lifespan below names its basis. A model ID date is a build date and can precede public availability. A bar with no retirement date in the sources stops at 2026-11-03. That does not mean the model is still served. 96 of 158 model families have no release date in the sources and are not drawn.
Retired models. OpenAI’s deprecations page S14 records the shutdown of o1-preview (2025-07-28) and o1-mini (2025-10-27). Their model IDs S08 date both releases to 2024-09-12, so o1-preview lasted 319 days derived and o1-mini 410 days derived. Anthropic’s deprecations page S55 lists retirement of Claude Instant (2024-11-06) and Claude 2 and 2.1 (2025-07-21). The same page S55 lists Claude 3 Opus as retired on 2026-01-05, 672 days derived after the launch that Anthropic’s post S60 dates to 2024-03-04. It lists Claude 3 Haiku as retired on 2026-04-20, after 774 days derived, counted from the build date in its model ID S55, 2024-03-07, not from a launch date. Anthropic’s launch post S60 says only that Haiku “will be available soon”, so no launch date is used for Haiku. OpenAI’s deprecations page S14 now says that access to twelve models and snapshots “was shut down on October 23, 2026”. They include gpt-4, gpt-4-turbo, gpt-3.5-turbo, gpt-4.1-nano, o1, o1-pro, o3-mini and o4-mini. The lifespans start at the DevDay post S57 for gpt-4-turbo and at model ID dates S14 for the others. They are 1,082 days derived for gpt-4-turbo, 557 days derived for gpt-4.1-nano and 675 days derived for o1. They are 583 days derived for o1-pro, 630 days derived for o3-mini and 555 days derived for o4-mini.
Google. Google’s deprecations page S56 gives Gemini 3 Pro Preview a release date of 2025-11-18 and a shutdown date of 2026-03-09. That is 111 days derived, computed as shutdown date minus release date. Google states that the shutdown dates it lists are the earliest possible dates, so the actual lifespan may be longer. On the same basis, the release date column of that page, Gemini 2.0 Flash ran from 2025-02-05 to 2026-06-01, 481 days derived.
Mistral AI and DeepSeek. Mistral AI’s model table S33 retired Mistral Large 2 (24.07, on 2025-03-30) and the first Mistral Large (24.02, on 2025-06-16). DeepSeek’s change log S38 announced that the names deepseek-chat and deepseek-reasoner would be discontinued on 2026-07-24. No later entry confirms the shutdown, and the live page no longer lists either name.
Scheduled retirements. OpenAI’s deprecations page S14 sets 2026-12-11 for GPT-5, GPT-5 mini, GPT-5 nano and o3, and says preview models may be retired with much shorter notice, such as 2 weeks. A notice on the same page S14, dated 2026-10-14, removes five snapshots from the API on 2027-01-14, with gpt-6-luna as the recommended replacement. They are gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14. Google’s deprecations page S56 lists 2026-12-01 as the shutdown date of gemini-2.5-pro and gemini-2.5-flash. It says requests to them route to gemini-3.8-flash and gemini-3.5-flash-lite from that date, and that its shutdown dates are the earliest possible dates. Mistral’s model table S33 lists Mistral Large 3 and Mistral Medium 3.5 as deprecated from 2026-10-31, with retirement on 2027-04-30. Anthropic’s page S55 schedules Claude Sonnet 4.5 to retire on 2026-11-30 and promises at least 60 days’ notice for publicly released models.
How to read these prices
Units. All prices are US dollars per 1M tokens at the standard pay-as-you-go rate for short prompts. OpenAI’s pricing page S04 still priced per 1K tokens on 2023-11-14 (dated), and this page multiplies such prices by 1,000. The first per-1M capture, of 2024-03-01 (dated), lists gpt-4 at $30.00 S05 per 1M input tokens.
Discounted tiers are not charted. Batch, cached, priority, long-context and off-peak rates are in the dataset only. On 2024-08-31 (dated) gpt-4o listed at $2.50 S07 per 1M input tokens in the Batch API against $5.00 S07 standard.
Snapshot dates. A change is dated to the first snapshot that shows it, not the day it took effect. Mistral AI has no readable snapshot between 2025-05-31 and 2026-10-07, so the date Mistral Large fell to $0.50 S33 is unknown.
Aliases. Some model names changed price with no change of name. gpt-4o pointed at the $5.00 S06 May 2024 model on 2024-05-31 (dated) and at the $2.50 S08 August model on 2024-12-14. DeepSeek’s deepseek-chat moved from DeepSeek-V2 to V3, V3.1 and V3.2 across four snapshots.
Host prices are excluded. Azure OpenAI, Amazon Bedrock, Together AI, Fireworks and OpenRouter prices are reseller prices and are not published. Meta publishes no API price, so Llama models do not appear. EUR and CNY prices are kept as written, with no exchange rate applied.
Tokens are not text. Tokenizers differ, so per-token prices are not per-text prices. Anthropic states S15 that its tokenizer for Claude 4.7 and later produces more tokens for the same text.
Methodology
Every price was copied from a vendor pricing page, docs table or announcement post, or from a Wayback Machine capture of one. Snapshots were retrieved on 2026-10-07, with two exceptions. The OpenAI live table rows were read on 2026-10-09 from the archive capture of the page as it stood on 2026-10-07. DeepSeek’s change log was retrieved on 2026-10-09. A coverage pass on 2026-10-09 re-read the saved pages and added every row of collapsed and tabbed tables. On 2026-10-20 and again on 2026-11-03 the live pricing pages of the six vendors and the deprecations pages of OpenAI, Anthropic and Google were read again. Prices that are new or changed have rows dated 2026-10-20 or 2026-11-03. The 2026-Q4 tier snapshot carries forward the latest earlier standard prices of models whose page showed no change, after checking each against the 2026-11-03 page text. The clean dataset holds 2,304 rows. Of these, 2,118 are publishable: 1,856 vendor list prices, 17 vendor announcement figures, 2 vendor retirement notice periods and 243 derived rows. Host prices, secondary analyses and xAI registry values with no stated unit are excluded.
Dated figures. The last full data check was on 2026-11-03, so the cut-off is 2024-11-03, 24 months earlier. A figure is labelled “dated” when the date that counts for it is before the cut-off. For a price on a live pricing page that date is the day it was read. For an archived snapshot it is the snapshot date, and for an announcement post its publication date. For a derived figure it is the earliest date it uses. Release and retirement dates, and the lifespans computed from them, are lifecycle dates and are never dated. The dataset’s dated column uses the same cut-off. Dated figures are historical records shown with their dates, not current prices.
Prices are reproduced as written. Derived percentages are rounded to one decimal place and lifespans are whole days. Each quarter-end point is the model with the lowest standard input price in the tier among models open to new customers, on the last snapshot of the quarter. Ties are kept and named. 2026-Q3 uses the live pages of 2026-10-07. 2026-Q4 is quarter to date and uses the live pages of 2026-11-03, so a later snapshot in the quarter replaces it. The GPT-4 comparison percentages are computed on the 2026-10-07 list prices, which are unchanged on 2026-10-20 and 2026-11-03. A model’s availability is read only from the vendor’s deprecations page or change log in the ledger. It is marked unknown where no such page covers the date.
The data does not capture every price change. OpenAI has no capture for 2025-Q1 or 2026-Q1. Anthropic has none between 2025-10-04 and 2026-03-02, and Mistral AI none between 2025-05-31 and 2026-10-07. The CSV download gives the tier of every observation (the tier column), the family of each model ID (model_family) and every snapshot date (observed_on). Every gap between snapshots can be read from it. The tier rules are in “How the tiers are defined” above. Hub-wide rules are on the methodology page.
Frequently asked questions
How much does an LLM API cost?
As of 2026-11-03, frontier-tier models cost $0.70 S33 to $10 S01 per 1M input tokens and $2.10 S33 to $50 S01 per 1M output tokens at standard list prices. Small-tier models cost $0.10 S01 to $0.75 S01 input and $0.50 S01 to $4.50 S01 output.
Which LLM has the cheapest API?
As of 2026-11-03, the lowest standard input price on the six vendors' pages is $0.05 S01 per 1M tokens, for OpenAI gpt-5-nano. OpenAI's deprecations page S14 schedules gpt-5-nano for removal on 2026-12-11. Among models open to new customers with no announced deprecation, gpt-6-luna lists at $0.10 S01 input per 1M tokens. Claude Haiku 5.5 lists at $0.10 S15 for prompts up to 100,000 tokens and Mistral's Ministral 3 3B at $0.10 S33, a three-way tie. Gemini 2.5 Flash-Lite also lists at $0.10 S25, but Google's deprecations page S56 limits access to the 2.5 models to existing users.
Is OpenAI API cheaper than Claude?
As of 2026-11-03, OpenAI's cheapest frontier-tier model, gpt-5.2, lists at $1.75 S01 input and $14.00 S01 output per 1M tokens, against $4 S15 and $20 S15 for Claude Opus 5.5. In the small tier, gpt-6-luna lists at $0.10 S01 and $0.50 S01, equal to Claude Haiku 5.5 at $0.10 S15 and $0.50 S15 for prompts up to 100,000 tokens.
Which LLM API is totally free?
None of the standard list prices recorded for the six vendors is zero; free tiers and their limits are outside this dataset. As of 2026-11-03, the lowest standard input price is $0.05 S01 per 1M tokens, for OpenAI gpt-5-nano.
Which is the best LLM API?
This dataset records list prices, not output quality, so it does not rank models. On price alone, as of 2026-11-03 the cheapest frontier-tier input is $0.70 S33 per 1M tokens for Mistral Large 4. The cheapest small-tier input open to new customers is $0.10 for gpt-6-luna S01 and for Claude Haiku 5.5 S15 up to 100,000 tokens, a tie. Gemini 2.5 Flash-Lite also lists at $0.10 S25, but Google limits access to existing users S56.
How much have LLM API prices dropped since GPT-4?
OpenAI listed GPT-4 (8K context) at $30 S02 input and $60 S02 output per 1M tokens on 2023-03-15 (dated). As of 2026-11-03, its cheapest frontier-tier model, gpt-5.2, lists at $1.75 S01 and $14.00 S01. That is 94.2% derived lower on input and 76.7% derived on output (both dated comparisons), computed as (old minus new) divided by old.
Sources
Every figure on this page carries a key that links to an entry below. Retrieved is the date the figure was read from the source for this dataset; the methodology section gives the date of the last full check. Sponsor names who paid for the research; a figure with a sponsor is read with that interest in mind. Keys follow the source ledger, so a missing number is a source that was reviewed and not used on this page. Hub-wide rules are on the methodology page.
S01 OpenAI, Pricing | OpenAI API (live page) (developers.openai.com)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
gpt-5.2: $1.75 input and $14.00 output per 1M tokens, standard tier. Derived against GPT-4 (8K context, S02): (30 minus 1.75) divided by 30 = 94.2% lower on input, (60 minus 14) divided by 60 = 76.7% lower on output. The row shows only under All models, so the quote is the table data in the page markup
- Verbatim
[[0,"gpt-5.2"],[0,1.75],[0,0.175],[0,14]]
gpt-6-luna: $0.10 input and $0.50 output per 1M tokens, short context
- Verbatim
gpt-6-luna | $0.10 | $0.50 | $0.01 | $0.125 | $0.25
gpt-5 and gpt-5.1: $1.25 per 1M input tokens. The row shows only under All models, so the quote is the table data in the page markup
- Verbatim
[[0,"gpt-5.1"],[0,1.25],[0,0.125],[0,10]] ... [[0,"gpt-5"],[0,1.25],[0,0.125],[0,10]]
gpt-5.3-codex: $1.75 per 1M input tokens, a specialised model
- Verbatim
Codex | gpt-5.3-codex | $1.75 | $14.00 | $0.175
gpt-6.1-sol: $2.00 input and $10.00 output per 1M tokens. Derived against GPT-4 (8K context, S02): (30 minus 2) divided by 30 = 93.3% lower on input, (60 minus 10) divided by 60 = 83.3% lower on output
- Verbatim
gpt-6.1-sol | $2.00 | $10.00 | $0.10 | $2.50 | $5.00
gpt-5-nano: $0.05 per 1M input tokens. The row shows only under All models, so the quote is the table data in the page markup
- Verbatim
[[0,"gpt-5-nano"],[0,0.05],[0,0.005],[0,0.4]]
gpt-5.6-sol: promotional $4.00 input and $20.00 output per 1M tokens, available at least through 2026-11-21
- Verbatim
GPT-5.6 Sol’s promotional pricing is available at least through November 21, 2026. ... gpt-5.6-sol | $4.00 | $20.00 | $0.40 | $5.00 | $10.00
Top of the frontier-tier range: gpt-6-astra at $10 input and $50 output per 1M tokens
- Verbatim
gpt-6-astra | $10.00 | $50.00 | $1.00 | $12.50 | $25.00
Top of the small-tier range: gpt-5.4-mini at $0.75 input and $4.50 output per 1M tokens. The row shows only under All models, so the quote is the table data in the page markup
- Verbatim
[[0,"gpt-5.4-mini"],[0,0.75],[0,0.075],[0,4.5]]
gpt-6.1-terra: $1.00 input and $5.00 output per 1M tokens on 2026-11-03, not on the 2026-10-20 page; a terra model, OpenAI's mid tier
- Verbatim
gpt-6.1-terra | $1.00 | $5.00 | $0.05 | $1.25 | $2.50
gpt-6-astra-law: promotional $10.00 input and $50.00 output per 1M tokens, available at least through 2026-12-31, on 2026-11-03
- Verbatim
gpt-6-astra-law | $10.00 | $50.00 | $1.00 | GPT-6 Astra for Law promotional pricing is available at least through December 31, 2026.
S02 OpenAI (Wayback Machine capture), Pricing, openai.com/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2023-03-15
- Retrieved
- 2026-10-07
GPT-4 (8K context): $0.03 input and $0.06 output per 1K tokens, or $30 and $60 per 1M tokens, 2023-03-15 (dated). The base of the derived 94.2%, 76.7%, 93.3% and 83.3% comparisons (formulas under S01)
- Verbatim
GPT-4 ... 8K context | $0.03 / 1K tokens | $0.06 / 1K tokens
gpt-3.5-turbo: one combined price of $0.002 per 1K tokens, or $2 per 1M tokens, 2023-03-15 (dated)
- Verbatim
gpt-3.5-turbo | $0.002 / 1K tokens
S04 OpenAI (Wayback Machine capture), Pricing, openai.com/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2023-11-14
- Retrieved
- 2026-10-07
GPT-4 Turbo (gpt-4-1106-preview): $0.01 per 1K input tokens, or $10 per 1M tokens, 2023-11-14 (dated); the page still priced per 1K tokens
- Verbatim
gpt-4-1106-preview | $0.01 / 1K tokens | $0.03 / 1K tokens
S05 OpenAI (Wayback Machine capture), Pricing, openai.com/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-03-01
- Retrieved
- 2026-10-07
gpt-3.5-turbo-0125: $0.50 per 1M input tokens, 2024-03-01 (dated)
- Verbatim
gpt-3.5-turbo-0125 | $0.50 / 1 M tokens | $1.50 / 1 M tokens
gpt-4: $30.00 per 1M input tokens in the first per-1M capture, 2024-03-01 (dated)
- Verbatim
gpt-4 | $30.00 / 1 M tokens | $60.00 / 1 M tokens
S06 OpenAI (Wayback Machine capture), Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-05-31
- Retrieved
- 2026-10-07
gpt-4o (the May 2024 model): $5.00 per 1M input tokens, 2024-05-31 (dated)
- Verbatim
gpt-4o | $5.00 / 1 M tokens | $15.00 / 1 M tokens
S07 OpenAI (Wayback Machine capture), Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-08-31
- Retrieved
- 2026-10-07
gpt-4o alias: $5.00 per 1M input tokens standard and $2.50 in the Batch API, 2024-08-31 (dated)
- Verbatim
gpt-4o | $5.00 / 1M input tokens | $2.50 / 1M input tokens
gpt-4o-2024-08-06, the August gpt-4o snapshot: $2.50 per 1M input tokens, 2024-08-31 (dated)
- Verbatim
gpt-4o-2024-08-06 | $2.50 / 1M input tokens
gpt-4o-mini: $0.150 per 1M input tokens, 2024-08-31 (dated)
- Verbatim
gpt-4o-mini | $0.150 / 1M input tokens
S08 OpenAI (Wayback Machine capture), Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-12-14
- Retrieved
- 2026-10-07
gpt-4o alias, pointing at the August model: $2.50 per 1M input tokens, 2024-12-14
- Verbatim
gpt-4o | $2.50 / 1M input tokens
Model IDs date o1-preview and o1-mini to 2024-09-12. Lifespans with the S14 shutdown dates: 2025-07-28 minus 2024-09-12 = 319 days (o1-preview), 2025-10-27 minus 2024-09-12 = 410 days (o1-mini)
- Verbatim
o1-preview-2024-09-12 ... o1-mini-2024-09-12
S09 OpenAI (Wayback Machine capture), API Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-04-19
- Retrieved
- 2026-10-07
OpenAI o3: $10.00 input and $40.00 output per 1M tokens, 2025-04-19. Start of the derived 80.0% cut: (10 minus 2) divided by 10 on input, (40 minus 8) divided by 40 on output
- Verbatim
OpenAI o3 ... Input: $10.00 / 1M tokens | Cached input: $2.50 / 1M tokens | Output: $40.00 / 1M tokens
GPT-4.1: $2.00 input and $8.00 output per 1M tokens, 2025-04-19
- Verbatim
GPT-4.1 ... Input: $2.00 / 1M tokens | Cached input: $0.50 / 1M tokens | Output: $8.00 / 1M tokens
S10 OpenAI (Wayback Machine capture), API Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-07-02
- Retrieved
- 2026-10-07
OpenAI o3: $2.00 input and $8.00 output per 1M tokens, 2025-07-02. End of the derived 80.0% cut (formula under S09)
- Verbatim
OpenAI o3 ... Input: $2.00 / 1M tokens | Cached input: $0.50 / 1M tokens | Output: $8.00 / 1M tokens
S11 OpenAI (Wayback Machine capture), API Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-08-31
- Retrieved
- 2026-10-07
GPT-5: $1.250 input and $10.000 output per 1M tokens, 2025-08-31
- Verbatim
GPT-5 ... Input: $1.250 / 1M tokens | Cached input: $0.125 / 1M tokens | Output: $10.000 / 1M tokens
GPT-5 nano: $0.050 per 1M input tokens, 2025-08-31
- Verbatim
GPT-5 nano ... Input: $0.050 / 1M tokens
GPT-5 mini: $0.250 per 1M input tokens, 2025-08-31
- Verbatim
GPT-5 mini ... Input: $0.250 / 1M tokens
S12 OpenAI (Wayback Machine capture), API Pricing, openai.com/api/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-12-14
- Retrieved
- 2026-10-07
GPT-5.2: $1.750 input and $14.000 output per 1M tokens, 2025-12-14
- Verbatim
GPT-5.2 ... Input: $1.750 / 1M tokens | Cached input: $0.175 / 1M tokens | Output: $14.000 / 1M tokens
GPT-5 mini: $2.000 per 1M output tokens, 2025-12-14
- Verbatim
GPT-5 mini ... Input: $0.250 / 1M tokens | Cached input: $0.025 / 1M tokens | Output: $2.000 / 1M tokens
S13 OpenAI (Wayback Machine capture), OpenAI API Pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2026-06-01
- Retrieved
- 2026-10-07
GPT-5.4: $15.00 per 1M output tokens, 2026-06-01
- Verbatim
GPT-5.4 ... Input: $2.50 / 1M tokens | Cached input: $0.25 / 1M tokens | Output: $15.00 / 1M tokens
GPT-5.5: $5.00 input and $30.00 output per 1M tokens, 2026-06-01
- Verbatim
GPT-5.5 ... Input: $5.00 / 1M tokens | Cached input: $0.50 / 1M tokens | Output: $30.00 / 1M tokens
GPT-5.4 mini: $0.75 per 1M input tokens, 2026-06-01
- Verbatim
GPT-5.4 mini ... Input: $0.75 / 1M tokens
S14 OpenAI, Deprecations | OpenAI API (live page) (developers.openai.com)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Removal of gpt-5 and gpt-5.1 announced in the entries dated 2026-06-11 and 2026-10-01 (headings, text and table cells)
- Verbatim
2026-10-01: GPT-5.3-Codex, GPT-5.1, GPT-5.4-Nano ... The following models are deprecated and will be removed from the API on April 1, 2027 ... Apr 1, 2027 | gpt-5.1 | gpt-5.1-2025-11-13 | gpt-6-sol ... 2026-06-11: GPT-5 and o3 model deprecations ... Dec 11, 2026 | gpt-5 | gpt-5-2025-08-07 | gpt-5.6-sol
Shutdown of o1-preview on 2025-07-28 and of o1-mini on 2025-10-27
- Verbatim
2025-04-28: o1-preview and o1-mini ... Shutdown date | Model / system | Recommended replacement | 2025-07-28 | o1-preview | o3 | 2025-10-27 | o1-mini | o4-mini
Removal on 2026-12-11 of GPT-5, GPT-5 mini, GPT-5 nano (gpt-5-nano) and o3 (text and table cells)
- Verbatim
On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. ... Dec 11, 2026 | gpt-5-mini | gpt-5-mini-2025-08-07 ... Dec 11, 2026 | gpt-5-nano | gpt-5-nano-2025-08-07 ... Dec 11, 2026 | o3 | o3-2025-04-16
Access to twelve models and snapshots, including gpt-4, gpt-4-turbo and o1, was shut down on 2026-10-23 (text and table cells)
- Verbatim
Access to these models was shut down on October 23, 2026. ... October 23, 2026 | gpt-3.5-turbo-0125 | gpt-3.5-turbo ... October 23, 2026 | gpt-4-0613 | gpt-4 ... October 23, 2026 | gpt-4-turbo | gpt-4-turbo-2024-04-09 ... October 23, 2026 | gpt-4.1-nano ... October 23, 2026 | o1-2024-12-17 | o1 ... October 23, 2026 | o3-mini-2025-01-31 | o3-mini ... October 23, 2026 | o4-mini-2025-04-16 | o4-mini
Preview models may be retired with much shorter notice, such as 2 weeks
- Verbatim
Preview models ... may be retired with much shorter notice, such as 2 weeks
Notice of 2026-10-14: five GPT-4 series snapshots in four rows (gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, gpt-4.1-mini-2025-04-14) leave the API on 2027-01-14, with gpt-6-luna as the replacement (text and table cells)
- Verbatim
On October 14, 2026, we notified developers using certain GPT-4 series models of their deprecation and removal from the API on January 14, 2027. ... January 14, 2027 | GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20) | gpt-6-luna ... January 14, 2027 | GPT-4o mini (gpt-4o-mini-2024-07-18) ... January 14, 2027 | GPT-4.1 (gpt-4.1-2025-04-14) ... January 14, 2027 | GPT-4.1 mini (gpt-4.1-mini-2025-04-14)
S15 Anthropic, Pricing (Claude Platform Docs, live page); legacy models section from claude.com/pricing (platform.claude.com)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Claude Opus 5.5: $4 input and $20 output per 1M tokens
- Verbatim
Claude Opus 5.5 ... $4 / MTok | $20 / MTok
Claude Haiku 5.5: $0.10 input and $0.50 output per 1M tokens for prompts up to 100,000 tokens; $0.50 and $2.50 above that. Not on the 2026-10-07 snapshot, listed on the 2026-10-20 and 2026-11-03 snapshots
- Verbatim
Claude Haiku 5.5 ... $0.10 / MTok for prompts up to 100,000 tokens | $0.50 / MTok ... $0.50 / MTok for prompts over 100,000 tokens | $2.50 / MTok
Claude Haiku 4.5: $1 input and $5 output per 1M tokens in the current model table on 2026-10-07
- Verbatim
Claude Haiku 4.5 | $1 / MTok | $5 / MTok
Claude Haiku 4.5 under Legacy models on 2026-10-20 and 2026-11-03, price unchanged
- Verbatim
Legacy models ... Haiku 4.5 Prompt caching Read $0.10 / MTok Write $1.25 / MTok Input $1 / MTok Output $5 / MTok
Claude Opus 4.1: $15 input and $75 output per 1M tokens, legacy table
- Verbatim
Claude Opus 4.1 | $15 / MTok | $75 / MTok
The tokenizer of Claude 4.7 and later produces more tokens for the same text
- Verbatim
Claude 4.7 and later models and Claude Mythos Preview use a newer tokenizer ... This tokenizer produces approximately 30% more tokens for the same text.
S16 Anthropic (Wayback Machine capture), Model Pricing (model_pricing_july2023.pdf) (web.archive.org)
- Sponsor
- none stated
- Published
- 2023-09-13
- Retrieved
- 2026-10-07
Claude 2: $11.02 input and $32.68 output per 1M tokens, price sheet archived 2023-09-13 (dated)
- Verbatim
Claude 2 | 100,000 tokens | Prompt $11.02/million tokens | Completion $32.68/million tokens
Claude Instant: $1.63 input and $5.51 output per 1M tokens, 2023-09-13 (dated). Start of the derived 56.4% output cut: (5.51 minus 2.40) divided by 5.51
- Verbatim
Claude Instant | 100,000 tokens | Prompt $1.63/million tokens | Completion $5.51/million tokens
S17 Anthropic (Wayback Machine capture), Claude API, anthropic.com/api (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-03-04
- Retrieved
- 2026-10-07
Claude 3 Opus: $15 input and $75 output per 1M tokens, 2024-03-04 (dated)
- Verbatim
Powerful Opus Input: $15 / MTok Output: $75 / MTok
Claude 2.0 and 2.1: $8 input and $24 output per 1M tokens, 2024-03-04 (dated)
- Verbatim
Claude 2.1 | 200,000 token context window | Input: $8 / MTok | Output: $24 / MTok ... Claude 2.0 | 100,000 token context window | Input: $8 / MTok | Output: $24 / MTok
Claude 3 Haiku: $0.25 input and $1.25 output per 1M tokens, 2024-03-04 (dated)
- Verbatim
Light & fast Haiku Input: $0.25 / MTok Output: $1.25 / MTok
Claude Instant: $2.40 per 1M output tokens, 2024-03-04 (dated). End of the derived 56.4% cut (formula under S16)
- Verbatim
Claude Instant | 100,000 token context window | Input: $0.80 / MTok | Output: $2.40 / MTok
S20 Anthropic (Wayback Machine capture), Models, docs.anthropic.com (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-11-20
- Retrieved
- 2026-10-07
Claude 3.5 Haiku: $1.00 input and $5.00 output per 1M tokens, 2024-11-20 (model comparison table cells)
- Verbatim
Claude 3.5 Sonnet | Claude 3.5 Haiku | Claude 3 Opus | Claude 3 Sonnet | Claude 3 Haiku ... Cost (Input / Output per MTok ) | $3.00 / $15.00 | $1.00 / $5.00
Claude 3 Haiku on the same page: $0.25 input and $1.25 output per 1M tokens (model comparison table cells)
- Verbatim
Claude 3.5 Sonnet | Claude 3.5 Haiku | Claude 3 Opus | Claude 3 Sonnet | Claude 3 Haiku ... Cost (Input / Output per MTok ) | $3.00 / $15.00 | $1.00 / $5.00 | $15.00 / $75.00 | $3.00 / $15.00 | $0.25 / $1.25
S22 Anthropic (Wayback Machine capture), Pricing, anthropic.com/pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-06-14
- Retrieved
- 2026-10-07
Claude Haiku 3 in the legacy section at $0.25 per 1M input tokens, 2025-06-14
- Verbatim
Claude Haiku 3 ... Input $0.25 / MTok
Claude Haiku 3.5: $0.80 input and $4 output per 1M tokens, 2025-06-14
- Verbatim
Claude Haiku 3.5 ... Input $0.80 / MTok | Prompt caching write $1 / MTok | Prompt caching read $0.08 / MTok | Output $4 / MTok
S23 Anthropic (Wayback Machine capture), Pricing | Claude (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-10-04
- Retrieved
- 2026-10-07
Opus 4.1: $15 input and $75 output per 1M tokens, 2025-10-04
- Verbatim
Opus 4.1 ... Input $15 / MTok | Output $75 / MTok
S24 Anthropic (Wayback Machine capture), Plans & Pricing | Claude by Anthropic (web.archive.org)
- Sponsor
- none stated
- Published
- 2026-03-02
- Retrieved
- 2026-10-07
Opus 4.6: $5 input and $25 output per 1M tokens for prompts up to 200K tokens, 2026-03-02
- Verbatim
Opus 4.6 Most intelligent model for building agents and coding Input Prompts ≤ 200K tokens $ 5 / MTok ... Output Prompts ≤ 200K tokens $ 25 / MTok
S25 Google, Gemini Developer API pricing (live page, now served at aistudio.google.com/docs/pricing) (ai.google.dev)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Gemini 3.1 Pro Preview: $2.00 input and $12.00 output per 1M tokens, prompts up to 200k tokens, on 2026-10-07, 2026-10-20 and 2026-11-03
- Verbatim
Gemini 3.1 Pro Preview ... Input price | Not available | $2.00, prompts <= 200k tokens ... Output price (including thinking tokens) | Not available | $12.00, prompts <= 200k tokens
Gemini 3.1 Flash-Lite: $0.25 input and $1.50 output per 1M tokens (price table cells)
- Verbatim
Gemini 3.1 Flash-Lite ... Input price | Free of charge | $0.25 (text / image / video) ... Output price (including thinking tokens) | Free of charge | $1.50
Gemini 2.5 Pro: $1.25 input and $10.00 output per 1M tokens, prompts up to 200k tokens (closed to new customers, S56)
- Verbatim
Gemini 2.5 Pro gemini-2.5-pro ... Input price | Free of charge | $1.25, prompts <= 200k tokens ... Output price (including thinking tokens) | Free of charge | $10.00, prompts <= 200k tokens
Gemini 2.5 Flash-Lite: $0.10 input and $0.40 output per 1M tokens (closed to new customers, S56)
- Verbatim
Gemini 2.5 Flash-Lite gemini-2.5-flash-lite ... Input price (text, image, video) | Free of charge | $0.10 (text / image / video) ... Output price (including thinking tokens) | Free of charge | $0.40
Gemini 4 Argon, not on the 2026-10-07 snapshot, listed on the 2026-10-20 and 2026-11-03 snapshots: $2.00 input and $10.00 output per 1M tokens through 2026-11-30; $4.00 and $20.00 from 2026-12-01
- Verbatim
Gemini 4 Argon gemini-4-argon ... Input price | Not available | $2.00 through November 30, 2026. $4.00 starting December 1, 2026. | Output price (including thinking tokens) | Not available | $10.00 through November 30, 2026. $20.00 starting December 1, 2026.
Google calls Gemini 4 Argon its flagship reasoning model; it counts as Google's frontier model by an editorial choice of 2026-10-22
- Verbatim
Gemini 4 Argon gemini-4-argon Try it in Google AI Studio Our flagship reasoning model
The 2026-11-03 copy of the page shows no update date (an absence, so there is no text to quote)
Gemini 3.6 Flash and 3.8 Flash (and 3.7 Flash on 2026-10-07): $0.75 input and $3.75 output per 1M tokens through 2026-12-31; $1.50 and $7.50 from 2027-01-01
- Verbatim
Gemini 3.8 Flash gemini-3.8-flash ... Input price | Free of charge | $0.75 through December 31, 2026. ... Gemini 3.6 Flash gemini-3.6-flash ... Input price | Free of charge | $0.75 through December 31, 2026. $1.50 starting January 1, 2027. | Output price (including thinking tokens) | Free of charge | $3.75 through December 31, 2026. $7.50 starting January 1, 2027.
Gemini 3.5 Flash and Gemini 3.7 Flash have no pricing block on the 2026-10-20 and 2026-11-03 pages (an absence, so there is no text to quote)
S26 Google (Wayback Machine capture), Gemini API Pricing | Google AI for Developers (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-05-30
- Retrieved
- 2026-10-07
Gemini 1.5 Flash: $0.35 input and $1.05 output per 1M tokens for prompts up to 128K tokens, 2024-05-30 (dated). Start of the derived cuts: (0.35 minus 0.075) divided by 0.35 = 78.6% on input, (1.05 minus 0.30) divided by 1.05 = 71.4% on output
- Verbatim
Gemini 1.5 Flash ... Price (input) | $0.35 / 1 million tokens (for prompts up to 128K tokens) ... Price (output) | $1.05 / 1 million tokens (for prompts up to 128K tokens)
Gemini 1.5 Pro: $3.50 input and $10.50 output per 1M tokens for prompts up to 128K tokens, 2024-05-30 (dated)
- Verbatim
Gemini 1.5 Pro ... Price (input) | $3.50 / 1 million tokens (for prompts up to 128K tokens) ... Price (output) | $10.50 / 1 million tokens (for prompts up to 128K tokens)
S27 Google (Wayback Machine capture), Gemini API Pricing | Google AI for Developers (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-08-29
- Retrieved
- 2026-10-07
Gemini 1.5 Flash: $0.075 input and $0.30 output per 1M tokens for prompts up to 128K tokens, 2024-08-29 (dated). End of the derived 78.6% and 71.4% cuts (formulas under S26)
- Verbatim
Gemini 1.5 Flash ... Prompts up to 128k tokens Input Pricing $0.075 / 1 million tokens output Pricing $0.30 / 1 million tokens
S28 Google (Wayback Machine capture), Gemini API pricing | Google AI for Developers (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-12-03
- Retrieved
- 2026-10-07
Gemini 1.5 Flash-8B: $0.0375 input and $0.15 output per 1M tokens for prompts up to 128K tokens, 2024-12-03; the lowest standard input price in the dataset
- Verbatim
Gemini 1.5 Flash-8B ... Prompts up to 128k tokens Input Pricing $0.0375 / 1 million tokens output Pricing $0.15 / 1 million tokens
Gemini 1.5 Pro: $1.25 input and $5.00 output per 1M tokens for prompts up to 128K tokens, 2024-12-03
- Verbatim
Gemini 1.5 Pro ... Prompts up to 128k tokens Input Pricing $1.25 / 1 million tokens output Pricing $5.00 / 1 million tokens
S29 Google (Wayback Machine capture), Gemini Developer API Pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-04-30
- Retrieved
- 2026-10-07
Gemini 2.5 Flash Preview: $0.15 per 1M input tokens; output $0.60 without thinking and $3.50 with thinking, 2025-04-30. Start of the derived 28.6% fall in thinking output: (3.50 minus 2.50) divided by 3.50
- Verbatim
Gemini 2.5 Flash Preview ... Input price | Free of charge | $0.15 (text / image / video) $1.00 (audio) | Output price | Free of charge | Non-thinking: $0.60 Thinking: $3.50
S30 Google (Wayback Machine capture), Gemini Developer API Pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-07-31
- Retrieved
- 2026-10-07
Gemini 2.5 Flash at general availability: $0.30 input and $2.50 output per 1M tokens, thinking tokens included, 2025-07-31. End of the derived 28.6% fall (formula under S29)
- Verbatim
Gemini 2.5 Flash ... Input price | Free of charge | $0.30 (text / image / video) $1.00 (audio) | Output price (including thinking tokens) | Free of charge | $2.50
S31 Google (Wayback Machine capture), Gemini Developer API pricing (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-12-31
- Retrieved
- 2026-10-07
Gemini 2.5 Pro: $1.25 input and $10.00 output per 1M tokens, prompts up to 200k tokens, 2025-12-31
- Verbatim
Gemini 2.5 Pro gemini-2.5-pro ... Input price | Free of charge | $1.25, prompts <= 200k tokens ... Output price (including thinking tokens) | Free of charge | $10.00, prompts <= 200k tokens
Gemini 2.0 Flash-Lite: $0.075 input and $0.30 output per 1M tokens, 2025-12-31 (price table cells)
- Verbatim
Gemini 2.0 Flash-Lite gemini-2.0-flash-lite ... Input price | Free of charge | $0.075 | Output price | Free of charge | $0.30
S33 Mistral AI, Pricing (Mistral AI docs); retirement dates from Models (docs.mistral.ai/models), retrieved 2026-11-03 (docs.mistral.ai)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Mistral Large 3: $0.50 input and $1.50 output per 1M tokens on 2026-10-07 and 2026-10-20 (the page writes $0.5 and $1.5); not on the 2026-11-03 pricing page
- Retrieved
- 2026-10-20
- Verbatim
Mistral Large 3 | $0.5 | $0.05 | $1.5
Mistral Small 4: $0.15 input and $0.60 output per 1M tokens
- Verbatim
Mistral Small 4 | $0.15 | $0.015 | $0.6
Mistral Large 4: $0.70 input and $2.10 output per 1M tokens on 2026-11-03 (the page writes $0.7 and $2.1), with no original or sale label
- Verbatim
Mistral Large 4 | $0.7 | $0.07 | $2.1
Mistral Large 4 on 2026-10-07 and 2026-10-20: original price $1.36 input and $4.18 output per 1M tokens; sale price $0.68 and $2.09, no end date shown
- Retrieved
- 2026-10-20
- Verbatim
Mistral Large 4 Sale price | Original price: $1.36 Sale price: $0.68 | Original price: $0.14 Sale price: $0.07 | Original price: $4.18 Sale price: $2.09
Mistral Large 3 and Mistral Medium 3.5: deprecation 2026-10-31, retirement 2027-04-30, alternative Mistral Large 4 (Mistral's model table, 2026-11-03)
- Verbatim
mistral-large-2512 ... 10/31/2026 ... 4/30/2027 ... Mistral Large 4
Retirement of Mistral Large 2 (24.07) on 2025-03-30 and of the first Mistral Large (24.02) on 2025-06-16
- Verbatim
Mistral Large 2.0 | 24.07 | mistral-large-2407 | 11/30/2024 3/30/2025 ... Mistral Large 1.0 | 24.02 | mistral-large-2402 | 11/30/2024 6/16/2025
Ministral 3 3B: $0.10 per 1M input tokens
- Verbatim
Ministral 3 3B | $0.1 | $0.01 | $0.1
S34 Mistral AI (Wayback Machine capture), Pricing and rate limits | Mistral AI Large Language Models (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-02-29
- Retrieved
- 2026-10-07
Mistral Large: $8 input and $24 output per 1M tokens, 2024-02-29 (dated). Start of the derived 75.0% input cut: (8 minus 2) divided by 8
- Verbatim
Mistral Large | mistral-large-latest | 8$ / 1M tokens | 24$ / 1M tokens
Mistral Small: $2 per 1M input tokens, 2024-02-29 (dated)
- Verbatim
Mistral Small | mistral-small-latest | 2$ / 1M tokens | 6$ / 1M tokens
S35 Mistral AI (Wayback Machine capture), Technology | Mistral AI (pricing section) (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-09-19
- Retrieved
- 2026-10-07
Mistral Large 2: $2 input and $6 output per 1M tokens, 2024-09-19 (dated)
- Verbatim
Mistral Large 2 | mistral-large-2407 ... $2 /1M tokens | $6 /1M tokens
Mistral Small 24.09: $0.2 per 1M input tokens, 2024-09-19 (dated)
- Verbatim
Mistral Small 24.09 | mistral-small-2409 ... $0.2 /1M tokens | $0.6 /1M tokens
Codestral: $0.2 per 1M input tokens, 2024-09-19 (dated)
- Verbatim
Codestral | codestral-2405 ... $0.2 /1M tokens | $0.6 /1M tokens
S36 Mistral AI (Wayback Machine capture), Pricing | Mistral AI (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-05-31
- Retrieved
- 2026-10-07
Mistral Small 3.1: $0.1 per 1M input tokens, 2025-05-31
- Verbatim
Mistral Small 3.1 ... Input (/M tokens) $0.1 | Output (/M tokens) $0.3
Mistral Large, no generation stated: $2 per 1M input tokens, 2025-05-31. End of the derived 75.0% cut (formula under S34)
- Verbatim
Mistral Large ... Input (/M tokens) $2 | Output (/M tokens) $6
Codestral: $0.3 per 1M input tokens, 2025-05-31
- Verbatim
Codestral ... Input (/M tokens) $0.3 | Output (/M tokens) $0.9
S38 DeepSeek, Models & Pricing (live page); discontinuation and off-peak dates from the Change Log (api-docs.deepseek.com/updates) (api-docs.deepseek.com)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
deepseek-v4-pro at the peak rate: $1.32 input (cache miss) and $3.96 output per 1M tokens (the table's columns are flash, then pro)
- Verbatim
1M INPUT TOKENS (CACHE MISS) | OFF-PEAK | $0.15 | $0.66 | PEAK | $0.3 | $1.32 ... 1M OUTPUT TOKENS | OFF-PEAK | $0.6 | $1.98 | PEAK | $1.2 | $3.96
deepseek-flash: $0.30 input (cache miss) and $1.20 output per 1M tokens at the peak rate; $0.15 input off-peak
- Verbatim
1M INPUT TOKENS (CACHE MISS) | OFF-PEAK | $0.15 | $0.66 | PEAK | $0.3 | $1.32 ... 1M OUTPUT TOKENS | OFF-PEAK | $0.6 | $1.98 | PEAK | $1.2 | $3.96
Off-peak rates are half the peak rates, in force from 16:00 UTC on 2026-08-16 (change log)
- Verbatim
we will adopt peak/off-peak pricing, with off-peak prices set at half of the peak-hour prices, encouraging users to schedule their tasks based on actual usage. The new prices will take effect at 16:00 (UTC Time) on August 16, 2026.
The names deepseek-chat and deepseek-reasoner to be discontinued on 2026-07-24 (change log)
- Verbatim
The two legacy API model names, deepseek-chat and deepseek-reasoner , will be discontinued in three months (2026-07-24).
S39 DeepSeek (Wayback Machine capture), Pricing | DeepSeek API Docs (web.archive.org)
- Sponsor
- none stated
- Published
- 2024-05-28
- Retrieved
- 2026-10-07
deepseek-chat: $0.14 input and $0.28 output per 1M tokens, 2024-05-28 (dated)
- Verbatim
deepseek-chat * | Good at general tasks, 32K * context length | $0.14 / 1M tokens | $0.28 / 1M tokens
S40 DeepSeek (Wayback Machine capture), Models & Pricing | DeepSeek API Docs (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-01-24
- Retrieved
- 2026-10-07
deepseek-chat list price $0.27 input and $1.10 output per 1M tokens, shown struck through beside a promotional $0.14 and $0.28 that applied until 16:00 UTC on 2025-02-08, 2025-01-24
- Verbatim
deepseek-chat | 64K | - | 8K | $0.07 (5) $0.014 | $0.27 (5) $0.14 | $1.10 (5) $0.28 ... From now until 2025-02-08 16:00 (UTC), all users can enjoy the discounted prices of DeepSeek API.
deepseek-reasoner: $0.55 input and $2.19 output per 1M tokens, 2025-01-24; not discounted
- Verbatim
deepseek-reasoner | 64K | 32K | 8K | $0.14 | $0.55 | $2.19 (6) ... DeepSeek-R1 is not included in the discount.
S41 DeepSeek (Wayback Machine capture), Models & Pricing | DeepSeek API Docs (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-09-10
- Retrieved
- 2026-10-07
deepseek-chat (one price for both aliases): $0.56 input and $1.68 output per 1M tokens, 2025-09-10. Start of the derived 75.0% output cut: (1.68 minus 0.42) divided by 1.68
- Verbatim
1M INPUT TOKENS (CACHE MISS) $0.56 | 1M OUTPUT TOKENS $1.68
S42 DeepSeek (Wayback Machine capture), Models & Pricing | DeepSeek API Docs (web.archive.org)
- Sponsor
- none stated
- Published
- 2026-03-01
- Retrieved
- 2026-10-07
deepseek-chat and deepseek-reasoner: $0.28 input and $0.42 output per 1M tokens, 2026-03-01. End of the derived 75.0% cut (formula under S41)
- Verbatim
1M INPUT TOKENS (CACHE MISS) $0.28 | 1M OUTPUT TOKENS $0.42
S43 xAI, Grok Models & Pricing (live page) (docs.x.ai)
- Sponsor
- none stated
- Published
- 2026-09-21
- Retrieved
- 2026-11-03
Grok 4.7: $2.00 input and $6.00 output per 1M tokens
- Verbatim
Grok 4.7 New grok-4.7 ... Context 500k tokens Input $2.00 / 1M tokens Output $6.00 / 1M tokens
S45 xAI (Wayback Machine capture), Models and Pricing (xAI documentation) (web.archive.org)
- Sponsor
- none stated
- Published
- 2025-09-11
- Retrieved
- 2026-10-07
grok-3: $3.00 input and $15.00 output per 1M tokens, 2025-09-11
- Verbatim
grok-3 | 131,072 | 600 rpm | $3.00 | $15.00
grok-3-mini: $0.30 input and $0.50 output per 1M tokens, 2025-09-11
- Verbatim
grok-3-mini | 131,072 | 480 rpm | $0.30 | $0.50
S55 Anthropic, Model deprecations, Claude Platform Docs (live page) (platform.claude.com)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Retirement of Claude Instant on 2024-11-06 and of Claude 2 and 2.1 on 2025-07-21 (text and table cells)
- Verbatim
These models were retired July 21, 2025. ... July 21, 2025 | claude-2.0 ... July 21, 2025 | claude-2.1 ... 2024-09-04: Claude 1 and Instant models ... These models were retired November 6, 2024. ... November 6, 2024 | claude-instant-1.2
Claude 3 Opus retired on 2026-01-05. Lifespan from the 2024-03-04 launch (S60): 2026-01-05 minus 2024-03-04 = 672 days
- Verbatim
2025-06-30: Claude Opus 3 model ... January 5, 2026 | claude-3-opus-20240229
Claude 3 Haiku retired on 2026-04-20. Lifespan from the build date in its model ID, 2024-03-07: 2026-04-20 minus 2024-03-07 = 774 days
- Verbatim
2026-02-19: Claude Haiku 3 model ... April 20, 2026 | claude-3-haiku-20240307
Claude Sonnet 4.5 to retire on 2026-11-30
- Verbatim
2026-09-30: Claude Sonnet 4.5 model ... November 30, 2026 | claude-sonnet-4-5-20250929
At least 60 days' notice before retirement for publicly released models
- Verbatim
Anthropic notifies customers ... providing at least 60 days' notice before model retirement for publicly released models.
S56 Google, Gemini deprecations (live page, now served at aistudio.google.com/docs/deprecations) (ai.google.dev)
- Sponsor
- none stated
- Published
- not stated
- Retrieved
- 2026-11-03
Access to the Gemini 2.5 models (2.5 Pro, 2.5 Flash, 2.5 Flash-Lite) limited to users who have used them before
- Verbatim
Access to the 2.5 models is limited to users who have actively used them in the past.
gemini-2.5-pro and gemini-2.5-flash are listed as deprecated on the 2026-11-03 copy (the 2026-10-20 copy said not deprecated), with shutdown date 2026-12-01 and routing to newer models from that date; gemini-2.5-flash-lite is not deprecated
- Verbatim
is deprecated and is no longer available to select in Google AI Studio. Access through the API is limited to existing users who have actively used it in the past. Effective December 1, 2026, all API requests to ... is deprecated and is no longer available to select in Google AI Studio. Effective December 1, 2026, all API requests to ... remains available through the API until further notice.
Gemini 3.5 Flash and 3.7 Flash: no shutdown date; the endpoints route API requests to Gemini 3.6 Flash and Gemini 3.8 Flash
- Verbatim
is no longer available in Google AI Studio; the endpoint is not shutting down and automatically routes API requests to
Gemini 3 Pro Preview: release 2025-11-18, shutdown 2026-03-09; 2026-03-09 minus 2025-11-18 = 111 days. Shutdown dates are the earliest possible dates
- Verbatim
The shutdown dates listed in the table indicate the earliest possible dates ... gemini-3-pro-preview | November 18, 2025 | March 9, 2026 | gemini-3.1-pro-preview
Gemini 2.0 Flash: release 2025-02-05, shutdown 2026-06-01; 2026-06-01 minus 2025-02-05 = 481 days
- Verbatim
gemini-2.0-flash | February 5, 2025 | June 1, 2026 | gemini-3.6-flash
S57 OpenAI (blog, Wayback Machine capture), New models and developer products announced at DevDay (web.archive.org)
- Sponsor
- none stated
- Published
- 2023-11-06
- Retrieved
- 2026-10-07
GPT-4 Turbo input at $0.01 per 1K tokens, DevDay post of 2023-11-06 (dated)
- Verbatim
GPT-4 Turbo input tokens are 3x cheaper than GPT-4 at $0.01 and output tokens are 2x cheaper at $0.03.
S58 Anthropic, Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku (anthropic.com)
- Sponsor
- none stated
- Published
- 2024-10-22
- Retrieved
- 2026-10-07
Claude 3.5 Haiku revised to $0.80 input and $4 output per 1M tokens on 2024-12-03
- Verbatim
Update (12/03/2024): We have revised the pricing for Claude 3.5 Haiku. The model is now priced at $0.80 MTok input / $4 MTok output.
S59 Google (Google for Developers blog), Updated production-ready Gemini models, reduced 1.5 Pro pricing, increased rate limits, and more (developers.googleblog.com)
- Sponsor
- none stated
- Published
- 2024-09-24
- Retrieved
- 2026-10-07
Gemini 1.5 Pro: 64% input and 52% output price cut, announced 2024-09-24 (dated), effective 2024-10-01
- Verbatim
Today we are announcing a 64% price reduction on input tokens, a 52% price reduction on output tokens, and a 64% price reduction on incremental cached tokens for our strongest 1.5 series model, Gemini 1.5 Pro, effective October 1st, 2024
S60 Anthropic, Introducing the next generation of Claude (anthropic.com)
- Sponsor
- none stated
- Published
- 2024-03-04
- Retrieved
- 2026-10-09
Claude 3 Opus and Sonnet available from the launch post of 2024-03-04, the start of the 672-day Opus lifespan (formula under S55)
- Verbatim
Opus and Sonnet are now available to use in claude.ai and the Claude API which is now generally available in 159 countries.
Claude 3 Haiku "will be available soon": the post gives no Haiku launch date
- Verbatim
Haiku will be available soon.
Figures marked derived are computed from the sourced figures beside them; the formula is stated in the text and rounding is stated once in the methodology.
Cite this page
Data on this page is published under CC BY 4.0. Use any figure, chart or table with a link back to this page.
- Plain text
Sysprobs Data (2026). LLM API Pricing Comparison 2026: Price History per 1M Tokens Since GPT-4. Edited by Dinesh Kumar. https://sysprobs.com/data/llm-api-price-history/ (updated 3 November 2026, accessed [date accessed]).- HTML
- APA
Sysprobs Data. (2026, November 3). LLM API Pricing Comparison 2026: Price History per 1M Tokens Since GPT-4. Retrieved [date accessed], from https://sysprobs.com/data/llm-api-price-history/
Quoting one figure: name the organisation that produced it, and its sponsor where there is one, then link here. For example: "As of 2026-11-03, the cheapest frontier-tier input price per vendor ranges from $0.70 per 1M tokens (Mistral AI's Mistral Large 4) to $4 per 1M tokens (Anthropic's Claude Opus 5.5).", compiled by Sysprobs Data. The source panel above gives the publisher, sponsor, sample and fieldwork for every key.
Changelog
- : First staged, data version 3.
- : Published at https://sysprobs.com/data/llm-api-price-history/, data version 5. List prices of six vendors from 2023-03-15 to 2026-11-03, read from vendor pricing pages and archived snapshots of them. Gemini 4 Argon counts as Google's frontier model (editorial choice of 2026-10-22).