What each AI lab charges for its current models, in US dollars per million tokens, input and output. Batch is the half-price queued tier where one is sold. “$ per call” is one call with 1,000 tokens in and 1,000 tokens out — a floor, since reasoning models write more than that. Epoch AI puts the fall in the price of a fixed level of capability at about 47% a quarter (The plunging price of thought, Sep 2026, CC BY 4.0).
| Lab | Model | Tier | Input $/M | Output $/M | Batch in / out | $ per call | Context | Listed | Hosts | Basis | Notes |
|---|---|---|---|---|---|---|---|---|---|---|---|
| OpenAI | GPT-6 Astra | flagship | 5.00 | 25.00 | 5.00 / 25.00 | 0.0300 | 1.05M | 2026-09-04 | 8 | lab endpoint | Astra Pro is the same price; time-of-day price up to $10 / $37.5 |
| OpenAI | GPT-6.1 Sol | mid | 1.00 | 5.00 | 1.00 / 5.00 | 0.0060 | 1.05M | 2026-09-29 | 9 | lab endpoint | Sol Pro is the same price; time-of-day price up to $2 / $7.5 |
| OpenAI | GPT-6.1 Terra | mid-low | 0.50 | 2.50 | 0.50 / 2.50 | 0.0030 | 1.05M | 2026-10-26 | 7 | lab endpoint | Terra Pro is the same price; time-of-day price up to $1 / $3.75 |
| OpenAI | GPT-6 Luna | small | 0.05 | 0.25 | 0.05 / 0.25 | 0.0003 | 1.05M | 2026-09-22 | 8 | lab endpoint | Luna Pro is the same price; the model behind Epoch AI's $0.0004-per-question example; time-of-day price up to $0.1 / $0.375 |
| Anthropic | Claude Fable 5.1 | flagship | 10.00 | 50.00 | 5.00 / 25.00 | 0.0600 | 1M | 2026-09-01 | 4 | lab endpoint | |
| Anthropic | Claude Opus 5.5 | upper | 4.00 | 20.00 | 2.00 / 10.00 | 0.0240 | 1M | 2026-09-22 | 11 | lab endpoint | |
| Anthropic | Claude Sonnet 5.5 | mid | 2.00 | 10.00 | 1.00 / 5.00 | 0.0120 | 1M | 2026-09-28 | 10 | lab endpoint | |
| Anthropic | Claude Haiku 5.5 | small | 0.10 | 0.50 | 0.05 / 0.25 | 0.0006 | 1M | 2026-10-07 | 10 | lab endpoint | time-of-day price up to $0.5 / $2.5 |
| Gemini 4 Argon | flagship | 2.00 | 10.00 | 1.00 / 5.00 | 0.0120 | 1.05M | 2026-10-13 | 6 | lab endpoint | ||
| Gemini 3.8 Flash | mid | 0.375 | 1.88 | 0.375 / 1.88 | 0.0022 | 1.05M | 2026-09-02 | 7 | lab endpoint | ||
| Gemini 3.5 Flash Lite | small | 0.15 | 1.25 | 0.15 / 1.25 | 0.0014 | 1.05M | 2026-07-21 | 8 | lab endpoint | ||
| xAI | Grok 4.7 | flagship | 2.00 | 6.00 | 1.60 / 4.80 | 0.0080 | 500K | 2026-09-21 | 6 | lab endpoint | No smaller Grok tier is sold in 2026; time-of-day price up to $4 / $12 |
| Meta | Muse Wonder | flagship | 4.00 | 20.00 | - | 0.0240 | 1.05M | 2026-10-12 | 2 | lab endpoint | |
| Meta | Muse Spark Open | mid | 1.25 | 4.25 | - | 0.0055 | 1.05M | 2026-10-15 | 4 | OpenRouter headline | Open weights; every host charges the same |
| Meta | Muse Glimmer 30B | small | 0.30 | 1.20 | 0.24 / 0.96 | 0.0015 | 131K | 2026-08-09 | 1 | OpenRouter headline | Open weights |
| DeepSeek | DeepSeek V4 Pro 0813 | flagship | 0.66 | 1.98 | - | 0.0026 | 1.05M | 2026-08-12 | 25 | lab endpoint | Open weights; DeepSeek's own endpoint shows the off-peak rate, peak hours (weekdays 01:00-04:00 and 06:00-10:00 UTC) are double; time-of-day price up to $1.32 / $3.96 |
| DeepSeek | DeepSeek V4.1 Flash 1017 | small | 0.15 | 0.60 | 0.011 / 0.20 | 0.0008 | 1.05M | 2026-10-17 | 51 | lab endpoint | Open weights; off-peak rate shown, peak hours are double; time-of-day price up to $0.3 / $1.2 |
| Mistral | Mistral Large 4 | flagship | 0.70 | 2.10 | - | 0.0028 | 1.05M | 2026-10-06 | 2 | lab endpoint | Launch discount ($0.68/$2.09) ended in October |
| Mistral | Mistral Medium 3.5 | mid | 1.50 | 7.50 | 0.75 / 3.75 | 0.0090 | 262K | 2026-04-30 | 2 | lab endpoint | |
| Mistral | Mistral Small 4 | small | 0.15 | 0.60 | - | 0.0008 | 262K | 2026-03-16 | 3 | lab endpoint | |
| Qwen (Alibaba) | Qwen3.8 Max Prime | flagship | 4.00 | 12.00 | - | 0.0160 | 1M | 2026-09-23 | 1 | lab endpoint | |
| Qwen (Alibaba) | Qwen3.8 Max (0902) | mid | 2.00 | 6.00 | 1.15 / 3.45 | 0.0080 | 1M | 2026-09-03 | 2 | lab endpoint | |
| Qwen (Alibaba) | Qwen3.8 Flash | small | 0.15 | 0.47 | - | 0.0006 | 1M | 2026-08-26 | 2 | lab endpoint | |
| Moonshot | Kimi K3.1 Pro | flagship | 2.00 | 10.00 | - | 0.0120 | 1.05M | 2026-10-23 | 23 | lab endpoint | |
| Moonshot | Kimi K3 | upper | 3.00 | 15.00 | 2.28 / 11.40 | 0.0180 | 1.05M | 2026-07-16 | 28 | lab endpoint | Open weights; 29 hosts |
| Z.ai | GLM 5.3 | flagship | 1.40 | 4.40 | 0.45 / 2.00 | 0.0058 | 1.05M | 2026-08-18 | 46 | lab endpoint | Open weights; 47 hosts |
| Z.ai | GLM 5.3 Flash | small | 0.15 | 0.50 | 0.042 / 0.061 | 0.0006 | 1.05M | 2026-08-26 | 42 | lab endpoint | Open weights; 41 hosts |
Where the numbers come from. OpenRouter’s public model list and per-host endpoint data (openrouter.ai), read once a day. A model’s price is the one charged by its own lab’s endpoint (25 of 27 rows), so a proprietary model shows its list price and an open-weight model shows what its lab charges, not whichever third-party host is cheapest that minute; the Hosts column counts the hosts and the CSV carries each model’s host range. List prices, not cost per task; prices change without notice. Not purchasing advice.
Every day, every model: prices-snapshots.csv · roster: prices\models.csv in the repository.