Model Capability Framework

Generated from GitHub Docs model data retrieved through gh api, then enriched with BaseCoat routing guidance. Effective user and organization entitlements must still be verified by the authenticated Copilot runtime.

Selection contract

1. Filter to GitHub-supported models and the target client.
2. Intersect with effective organization and user policy when runtime data is available.
3. Match task capability requirements before cost or latency.
4. Emit reasoning_effort only when the model is configurable and the authenticated runtime reports the exact effort value.
5. Prefer the lowest-cost model that passes task-specific evaluations.

reasoning_depth describes task complexity. It is not a provider parameter and must never be copied automatically into reasoning_effort.

Models

Model	Provider	Status	Copilot CLI	CLI auto-selection	Configurable reasoning	Task area
gpt-5-mini	OpenAI	GA	True	True	False	General-purpose coding and writing
kimi-k3	Moonshot AI	GA	True	False	True	Agentic coding and long-context work
kimi-k2.7-code	Moonshot AI	GA	True	False	False	General-purpose coding and agent tasks
mai-code-1.1-flash	Microsoft	GA	True	True	False	General-purpose coding and writing, image understanding
mai-code-1-flash-picker	Microsoft	GA	True	False	False	General-purpose coding and writing
gemini-3.8-flash	Google	GA	True	False	False	Fast help with simple or repetitive tasks
gemini-3.7-flash	Google	GA	True	True	False	Fast help with simple or repetitive tasks
gemini-3.6-flash	Google	GA	True	True	False	Fast help with simple or repetitive tasks
gemini-3.5-flash	Google	GA	True	False	False	Fast help with simple or repetitive tasks
claude-sonnet-5	Anthropic	GA	True	True	True	General-purpose coding and agent tasks
claude-sonnet-4.6	Anthropic	GA	True	True	True	General-purpose coding and agent tasks
claude-opus-5	Anthropic	GA	True	True	True	Deep reasoning and debugging
claude-opus-4.8-fast	Anthropic	GA	True	False	True	Deep reasoning and debugging
grok-4.5	xAI	GA	True	False	False	General-purpose coding and agent tasks
claude-opus-4.8	Anthropic	GA	True	True	True	Deep reasoning and debugging
claude-haiku-4.5	Anthropic	GA	True	True	False	Fast help with simple or repetitive tasks
claude-fable-5.1	Anthropic	GA	True	False	True	Long-horizon, autonomous coding and knowledge-work
claude-fable-5	Anthropic	GA	True	False	True	Long-horizon, autonomous coding and knowledge-work
gpt-6-astra	OpenAI	GA	True	False	True	Long-horizon, autonomous coding and agentic tasks
gpt-5.6-terra	OpenAI	GA	True	True	True	General-purpose coding and agent tasks
gpt-5.6-sol	OpenAI	GA	True	True	True	Deep reasoning and debugging
gpt-5.6-luna	OpenAI	GA	True	True	True	Fast help with simple or repetitive tasks
gpt-5.5	OpenAI	GA	True	True	True	Deep reasoning and debugging
gpt-5.4-nano	OpenAI	GA	False	False	False	Not classified by GitHub Docs
gpt-5.4-mini	OpenAI	GA	True	True	False	Agentic software development
gpt-5.4	OpenAI	GA	True	True	True	Deep reasoning and debugging
gpt-5.3-codex	OpenAI	GA	True	False	True	Agentic software development
claude-opus-4.7	Anthropic	GA	True	False	True	Deep reasoning and debugging
grok-4.6	xAI	GA	True	False	False	General-purpose coding and agent tasks

Runtime availability

GitHub public REST does not currently expose the effective per-user Copilot model allowlist. The refresh script therefore records the public supported baseline from github/docs; callers must intersect it with the access-filtered Copilot runtime catalog when available. Do not substitute the GitHub Models inference catalog.