Compare AI models
Price, context window and open weights for the models people actually choose between. Prices are the vendor's standard rate in US dollars per million tokens. Last checked 25 Oct 2026. Select a column to sort.
| Claude Haiku 5.5 | Anthropic | Oct 2026 | 1M | $0.10 | $0.50 | No | Fast, cheap high-volume work (price rises >100K) |
| GPT-6 Luna | OpenAI | Sept 2026 | 1.05M | $0.10 | $0.50 | No | Cheap, focused high-volume tasks |
| Mistral Small 4 | Mistral | Mar 2026 | 256K | $0.15 | $0.60 | Yes | Cheap Apache-2.0 MoE; API price shown |
| Gemini 3.5 Flash-Lite | Jul 2026 | 1.05M | $0.30 | $2.50 | No | Cheapest current Gemini for high throughput | |
| DeepSeek V4.1 Flash | DeepSeek | Oct 2026 | 1M | $0.30 | $1.20 | Yes | Very cheap open model; API peak price shown |
| Gemini 3.8 Flash | Sept 2026 | 1.05M | $0.75 | $3.75 | No | Fast agents and coding; price doubles Jan 2027 | |
| DeepSeek V4-Pro | DeepSeek | Aug 2026 | 1M | $1.32 | $3.96 | Yes | MIT-licensed frontier model; API peak price shown |
| Mistral Large 4 | Mistral | Oct 2026 | 1M | $1.36 | $4.18 | No | Public preview; open weights promised, not out yet |
| GPT-6.1 Sol | OpenAI | Sept 2026 | 1.05M | $2.00 | $10.00 | No | Near-Astra quality at a fifth of the price |
| Gemini 4 Argon | Oct 2026 | 1.05M | $2.00 | $10.00 | No | Flagship reasoning; price doubles Dec 1, 2026 | |
| Grok 4.7 | xAI | Sept 2026 | 500K | $2.00 | $6.00 | No | xAI flagship for code; only chat model listed |
| Claude Opus 5.5 | Anthropic | Sept 2026 | 1M | $4.00 | $20.00 | No | Agentic coding at lower cost than Fable |
| Claude Fable 5.1 | Anthropic | Sept 2026 | 1M | $10.00 | $50.00 | No | Hardest coding and knowledge work |
| GPT-6 Astra | OpenAI | Sept 2026 | 1.05M | $10.00 | $50.00 | No | OpenAI's most capable model for demanding work |
| Qwen3.8-2.4T-A95B | Alibaba Qwen | Aug 2026 | 262K | – | – | Yes | Largest open Qwen; self-hosted, custom license |
| Qwen3.8-27B | Alibaba Qwen | Aug 2026 | 262K | – | – | Yes | Apache-2.0 model that fits a single big GPU |
| Kimi K3 | Moonshot AI | Jul 2026 | 1.05M | – | – | Yes | 2.8T open MoE for agentic and long-context work |
A dash means the vendor does not publish that figure, or the model is mainly self-hosted. Each model name links to the page the figures were checked against.