What every major model costs per million tokens
Vendor list prices, read from the vendor's own pricing page, with the date Wren read them. Where a vendor publishes no simple list price, the blended figure is Artificial Analysis' estimate and is marked. Updated Oct 10, 2026.
| Model | Vendor | Input / 1M | Output / 1M | Blended / 1M | Context | Rank | Source |
|---|---|---|---|---|---|---|---|
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.50 | $0.20 | 1M | #7 | Claude pricing |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | $0.20 | 1M | #10 | OpenAI API pricing |
| DeepSeek Flash (4.1) | DeepSeek | $0.30 | $1.20 | $0.52 | 1M | #9 | DeepSeek API pricing |
| Muse Spark 1.3 | Meta | — | — | $0.78* | — | #8 | Artificial Analysis |
| Gemini 4 Argon | — | — | $1.99* | 1M | #3 | Artificial Analysis | |
| Kimi K3 | Moonshot | — | — | $2.31* | — | #11 | Artificial Analysis |
| Grok 4.7 | xAI | — | — | $3.74* | 500K | #12 | Artificial Analysis |
| Claude Sonnet 5.5 | Anthropic | $2 | $10 | $4 | 1M | #4 | Claude pricing |
| GPT-6.1 Sol | OpenAI | $2 | $10 | $4 | 1M | #5 | OpenAI API pricing |
| Claude Opus 5.5 | Anthropic | $4 | $20 | $8 | 1M | #1 | Claude pricing |
| GPT-6 Astra | OpenAI | $10 | $50 | $20 | 1M | #2 | OpenAI API pricing |
| Claude Fable 5.1 | Anthropic | $10 | $50 | $20 | 1M | #6 | Claude pricing |
* Artificial Analysis blended estimate; the vendor publishes no simple list price. Blended assumes three input tokens per output token.
Questions people ask
What is the cheapest major AI model per million tokens?
Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens on Anthropic's list price (October 2026).
What is the cheapest frontier model?
Gemini 4 Argon, at about $1.99 per million tokens blended (three input to one output), an Artificial Analysis estimate.
What is the most expensive model on the table?
Claude Fable 5.1, at about $20 per million tokens blended.
What does "blended price" mean?
A single per-million-token figure that assumes three input tokens for every output token, the convention Artificial Analysis uses. Real workloads differ: long prompts skew toward the input price, long generations toward the output price.
Prices change; the leaderboard and this table are refreshed by Wren every Monday and after major releases. See Breaking for price changes as they land.