
What every model costs per million tokens, at every provider that serves it. Frontier labs, open-weights hosts and the Chinese market, scraped daily, with history back to July 26, 2023.
The frontier (p90) has fallen 89% since September 30, 2023; the floor (p10) has barely moved. Deflation is replacement, not repricing: see the research charts →
What the labs themselves charge today, as of November 3, 2026
| O1 Pro OpenAI | $262.50 | 1 |
| GPT 5.4 Pro OpenAI | $67.50 | 2 |
| GPT 5.5 Pro OpenAI | $67.50 | 2 |
| GPT 5.2 Pro OpenAI | $57.75 | 2 |
| GPT 5 Pro OpenAI | $41.25 | 2 |
| GPT 4 OpenAI | $37.50 | 2 |
| O3 Pro OpenAI | $35.00 | 2 |
| Claude Opus 4.1 Anthropic | $30.00 | 2 |
| Claude Opus 4 Anthropic | $30.00 | 1 |
| O1 OpenAI | $26.25 | 1 |
| Claude Fable 5 Anthropic | $20.00 | 2 |
| Claude Fable 5.1 Anthropic | $20.00 | 2 |
| GPT-6 Astra OpenAI | $20.00 | 2 |
| GPT 4 Turbo OpenAI | $15.00 | 1 |
| GPT 5.5 OpenAI | $11.25 | 2 |
Open-weights models by number of providers listing them, and the cheapest of those listings
| GLM 5.3 Z.ai | $0.402 | 37 |
| GLM 5.3 Flash Z.ai | $0.0446 | 35 |
| DeepSeek V4.1 Flash 1017 DeepSeek | $0.105 | 34 |
| DeepSeek V4.1 Flash DeepSeek | $0.0578 | 31 |
| GLM 5.2 Z.ai | $0.415 | 24 |
| DeepSeek V4 Flash 0731 DeepSeek | $0.0875 | 24 |
| GPT OSS 120B OpenAI | $0.0600 | 23 |
| DeepSeek V4 Pro 0813 DeepSeek | $0.554 | 22 |
| Qwen3.8 27B Qwen | $0.342 | 22 |
| Kimi K3 Moonshot AI | $2.61 | 20 |
| Kimi K2.6 Moonshot AI | $0.702 | 17 |
| DeepSeek V4 Flash DeepSeek | $0.103 | 16 |
| DeepSeek V4 Pro DeepSeek | $1.05 | 14 |
| Gemma 4 31B IT Google | $0.160 | 14 |
| Gemma 4 26B A4B IT Google | $0.0675 | 14 |
| GLM 5.1 Z.ai | $1.38 | 13 |
| MiniMax M3.1 Flash MiniMax | $0.0875 | 13 |
| MiniMax M3 MiniMax | $0.388 | 12 |
| Kimi K2.7 Code Moonshot AI | $1.20 | 11 |
| Qwen3.5 397B A17B Qwen | $0.877 | 10 |
Repricings of 15% or more in the last 30 days, vendors only. Token prices move in steps, not drifts.
The same model is sold under a dozen names by a dozen hosts at a wide spread of prices. We resolve the names to one canonical model and show you the whole market for it, cheapest first.
Reconstructed from archived rate cards and daily API captures back to mid-2023, then observed live every day since. Every change point carries its source and a confidence grade.
DeepSeek, Qwen, Kimi, GLM, MiniMax and their hosts, priced in USD alongside the US labs. Nobody else tracks this side of the market with history.
Every observation is checked against plausibility bounds, cross-provider medians and yesterday's price before it is published. Failures are quarantined, not shown. Nothing on this site is a placeholder.
Start free. Upgrade when you need the whole matrix.
Paid plans open shortly. Until then, email us and we will set you up by hand.
The market at a glance
Every model, every provider
Three years of history, exportable
Engineers picking a host, analysts modelling the cost of intelligence, founders pricing a product: the number you need is here, with the history behind it.
Get started free