The price of GPT‑4‑level AI fell 682-fold in about three years.
In March 2023 the only model at that level was GPT-4 itself, at $37.50 per million tokens. By July 2026 it was Qwen3.7 Flash, at $0.055. The newest top models did not get cheaper the same way: OpenAI’s flagship still lists in the tens of dollars per million tokens.
Brass AI price index, updated Oct 1, 2026.
What $25 buys
At list price, $25 bought about 667,000 tokens of GPT-4-level work in March 2023 and about 455,000,000 in July 2026. Newer models often write more tokens per answer, so the cost of one correct answer on this test fell about 70-fold over the same period rather than 682-fold. A question cost about $0.031 on GPT-4 and $0.00044 on Qwen3.7 Flash with reasoning off, in the per-question cost data built on Epoch AI’s GPQA Diamond runs.
How we measure
- Capability: a model counts as GPT-4-level when its GPQA Diamond score is at least the original GPT-4’s, using Epoch AI’s benchmark data.
- Price: each developer’s own published list price on the date shown, checked against an archived copy of the pricing page or the launch post.
- Blend: three input tokens for every output token, which is close to how chat is used.
- Every row links to its sources and was checked against a primary source; the downloads carry a status column.
For what each model costs today, per million words and read from its maker’s pricing page every hour, see the market board.
Download the data
prices.csv and prices.json. Free to reuse under CC BY 4.0; credit “Brass AI price index, brass.adorellc.pro”. We add a row when a cheaper GPT-4-level model or a new OpenAI flagship ships.
The full table
US dollars per million tokens.
GPT-4-level work
| Date | Model | Input | Output | Blended | Sources |
|---|---|---|---|---|---|
| 2023-03-14 | GPT-4 (gpt-4-0314) | $30.00 | $60.00 | $37.50 | 1 17 |
| 2023-11-06 | GPT-4 Turbo (gpt-4-1106-preview) | $10.00 | $30.00 | $15.00 | 2 17 |
| 2024-02-26 | Mistral Large (mistral-large-2402) | $8.00 | $24.00 | $12.00 | 3 17 |
| 2024-03-04 | Claude 3 Sonnet | $3.00 | $15.00 | $6.00 | 4 17 |
| 2024-03-13 | Claude 3 Haiku | $0.25 | $1.25 | $0.50 | 5 17 |
| 2024-07-18 | GPT-4o mini | $0.15 | $0.60 | $0.2625 | 6 17 |
| 2024-08-12 | Gemini 1.5 Flash (price cut) | $0.075 | $0.30 | $0.1313 | 7 17 |
| 2025-01-30 | Qwen Turbo (qwen-turbo-2024-11-01) | $0.05 | $0.20 | $0.0875 | 8 17 |
| 2026-07-27 | Qwen3.7 Flash | $0.03 | $0.13 | $0.055 | 9 17 |
OpenAI flagship
| Date | Model | Input | Output | Blended | Sources |
|---|---|---|---|---|---|
| 2023-03-14 | GPT-4 | $30.00 | $60.00 | $37.50 | 1 |
| 2023-11-06 | GPT-4 Turbo | $10.00 | $30.00 | $15.00 | 2 |
| 2024-05-13 | GPT-4o | $5.00 | $15.00 | $7.50 | 10 |
| 2025-02-27 | GPT-4.5 preview | $75.00 | $150.00 | $93.75 | 11 |
| 2025-08-07 | GPT-5 | $1.25 | $10.00 | $3.4375 | 12 |
| 2025-12-11 | GPT-5.2 | $1.75 | $14.00 | $4.8125 | 13 |
| 2026-03-05 | GPT-5.4 | $2.50 | $15.00 | $5.625 | 14 |
| 2026-04-24 | GPT-5.5 | $5.00 | $30.00 | $11.25 | 15 |
| 2026-09-03 | GPT-6 Astra | $10.00 | $50.00 | $20.00 | 16 |
Sources
- OpenAI pricing page, archived 2023-03-15 (GPT-4 8K context: $0.03/1K prompt, $0.06/1K completion)
- OpenAI DevDay announcement, archived 2023-11-06 ($0.01/1K input, $0.03/1K output)
- Mistral pricing page, archived 2024-02-26 (Mistral Large $8/$24 per 1M)
- Anthropic, Introducing the next generation of Claude (Mar 4, 2024; Sonnet available in the API that day)
- Anthropic, Claude 3 Haiku release post (Mar 13, 2024); price from the Mar 4, 2024 Claude 3 announcement
- OpenAI, GPT-4o mini: advancing cost-efficient intelligence (Jul 18, 2024), archived
- Google Developers Blog, Gemini 1.5 Flash price drop effective Aug 12, 2024 (prompts under 128K), archived
- Alibaba Cloud Model Studio models page (international), archived 2025-01-30 (qwen-turbo $0.00005/1K input, $0.0002/1K output)
- Alibaba Cloud Model Studio pricing, International (Singapore) deployment, prompts up to 32K tokens
- OpenAI API pricing page, archived 2024-05-13 (gpt-4o-2024-05-13 $5/$15)
- OpenAI model page, GPT-4.5 Preview (snapshot gpt-4.5-preview-2025-02-27)
- OpenAI model page, GPT-5 (snapshot gpt-5-2025-08-07)
- OpenAI model page, GPT-5.2 (snapshot gpt-5.2-2025-12-11)
- OpenAI model page, GPT-5.4 (snapshot gpt-5.4-2026-03-05; prompts up to 272K)
- OpenAI model page, GPT-5.5 (prompts up to 272K); API release Apr 24, 2026 per OpenAI changelog
- OpenAI model page, GPT-6 Astra (prompts up to 272K); released Sep 3, 2026 per OpenAI changelog
- Epoch AI benchmark data, GPQA Diamond scores (downloaded 2026-10-01)