cmdop models

List available models for chat and skill execution. cmdop accepts both aliases (semantic, stable) and concrete IDs (specific provider model).

List


cmdop models
cmdop models --json

Example output:


ALIAS         CONCRETE                          PROVIDER     CONTEXT
@code         anthropic/claude-opus-4-7         sdkrouter    1M
@balanced     openai/gpt-4o                     sdkrouter    128K
@fast         openai/gpt-4o-mini                sdkrouter    128K
@vision       anthropic/claude-sonnet-4         sdkrouter    200K
@reasoning    z-ai/glm-5.1                      sdkrouter    256K

Aliases vs concrete IDs

Aliases (@code, @balanced, @fast, @vision, @reasoning) resolve via SDKRouter to whatever the team currently considers the best fit. Aliases survive provider migrations.
Concrete IDs (openai/gpt-4o-mini, z-ai/glm-5.1, anthropic/claude-opus-4-7) lock to one specific model. Use these in regression tests or when you want strict determinism.

Switch the model

Inside chat:


/model @code
/model anthropic/claude-opus-4-7

From the CLI:


cmdop chat --model @code
cmdop chat --model openai/gpt-4o-mini
cmdop run code-review src/ --model @code

Persist the choice:


cmdop config set chat.default_model @code

Custom providers

If you’ve added a custom OpenAI-compatible provider via cmdop provider add, that provider’s models also appear in the list — prefixed with the provider name:


cmdop provider set ollama-local
cmdop models
# → ollama-local/qwen2.5-coder:32b
# → ollama-local/llama3.3:70b

When a custom provider is active, SDKRouter aliases (@code, @balanced) resolve to that provider’s default model. Search, vision, and CDN still route through SDKRouter regardless of the chat provider. See ./provider.

Cost & latency hints

Alias	Best for	Latency	Relative cost
`@fast`	Short answers, simple lookups	Lowest	Lowest
`@balanced`	Day-to-day chat, refactors	Mid	Mid
`@code`	Heavy code work, long context	Higher	Higher
`@vision`	Image / screenshot tasks	Higher	Higher
`@reasoning`	Multi-step planning, debugging	Highest	Highest