Models
Quality tiers, prices, and what happens when a model is replaced.
Where the list comes from
The model list, the calculator, routing tiers, and bring-your-own-key pickers read one published catalog. It was last refreshed 2026-10-12 under policy latest-flagship-2026-10. A weekly job opens a pull request, runs the same tests as any other change, and squash-merges when they pass. Until that merge, the previous catalog stays in use. No Allow auto-merge setting is required.
Tiers
Economy starts near $0.0003 for a typical message. 1. Claude Haiku 5.5 ($0.10 in) 2. DeepSeek V4.1 Flash ($0.15 in) 3. Mistral Small 4 ($0.15 in) 4. Llama 4 Maverick ($0.1875 in) 5. Gemini 3.5 Flash Lite ($0.30 in) 6. GPT-5.4 Mini ($0.75 in) Balanced starts near $0.0013 for a typical message. 1. DeepSeek V4 Pro 0813 ($0.66 in) 2. Mistral Large 4 ($0.68 in) 3. Grok 4.20 ($1.25 in) 4. Gemini 3.8 Flash ($0.75 in) 5. Claude Sonnet 5.5 ($2 in) 6. GPT-5.4 ($2.50 in) Frontier starts near $0.0040 for a typical message. 1. Grok 4.7 ($2 in) 2. Gemini 3.1 Pro Preview ($2 in) 3. Claude Opus 5.5 ($4 in) 4. GPT-5.5 ($5 in)
What changed
2026-10-12: Published the OpenRouter snapshot under the latest-flagship policy. Not auto-placed: Claude Fable, GPT-5 chat, GPT-5 luna, GPT-5 luna-pro, GPT-5 pro, GPT-5 sol, GPT-5 sol-pro, GPT-5 terra, GPT-5 terra-pro.
When a model is replaced
The catalog records a successor when a tier model leaves the list. The id that left points at the model that replaced it, and the reason is either removed or promoted. That mapping is what the agent upgrade job reads. A model that is still in the snapshot stays callable.