CommandAGI · commandagi.com/for/any-model

CommandAGI
Open the workbench
Menu
Open the workbenchSign in

Any model

The catalog, or your own plans.

An agent names its model, and the platform routes the call to the provider that serves it. Every model in the catalog shows its price per million tokens before you pick it.

Open the workbenchSee harnesses and bases
Settings → agent / mcp in the CommandAGI workbench: the agent CLIs on this computer, how each one is signed in, and the plan of each sign-in.
Recorded in the workbench: the agent CLIs on this computer and the plan of each sign-in, beside the catalog's models.

One name, one route

An agent names a model, not a provider.

The agent file says model="anthropic/claude-sonnet-5-5". The catalog row says which provider serves it, through which API, and at what price.

How a model call is routedAn agent's file names a model by its catalog id. The catalog row names the model's provider, its API format, its price per million tokens in and out, its context and what it reads. The platform sends the call through that provider and meters it at that price. An agent CLI on your own plan is a route of its own and is not metered by the catalog.agent.tsxmodel="<provider>/<mod-el>"The catalog rowprovider · price · contextThe providerAnthropic · OpenAI ·DashScope · Fireworks …The meterat the row's priceHow a model call is routedAn agent's file names a model by its catalog id. The catalog row names the model's provider, its API format, its price per million tokens in and out, its context and what it reads. The platform sends the call through that provider and meters it at that price. An agent CLI on your own plan is a route of its own and is not metered by the catalog.agent.tsxmodel="<provider>/<model>"The catalog rowprovider · price · contextThe providerAnthropic · OpenAI · DashScope · Fireworks …The meterat the row's price

Case study · the production catalog

43 models, eight providers.

The catalog as CommandAGI's account read it on 2026-11-04 (case-studies/models/stats.json). Every model takes tools; 35 read images as well as text.

Your own keys. No model in the catalog runs on your own key yet (every row's byo is false). Your own subscriptions run through the agent CLIs instead: see them.

models, all enabled
43
providers: DashScope 11, OpenAI 10, Anthropic 8, Fireworks 7, OpenRouter 3, ComputaCo 2, CommandAGI 1, DeepInfra 1
8
per million output tokens
$0.30 to $50
tokens of context at the top
1 M
Price per million output tokens, a sample of the catalog
The numbers
Price per million output tokens, a sample of the catalog
ModelOutput, $ per million tokens
Claude Fable 550
GPT-6 Astra50
Claude Opus 5.520
Kimi K315
Claude Sonnet 5.510
GPT-6.1 Sol10
GLM-5.34.4
DeepSeek V4.1 Flash1.2
Claude Haiku 5.50.5
Llama 4 Scout0.3
ProvenanceFile GET /v1/models (production)Output case-studies/models/stats.jsonTool scripts/sites/case-studies.mjs modelsRan on This computer (dell-gb10-1), reading the production API as user-commandagiWhen

Pick a model per agent.

Give each agent the model its work needs, and see the price before it runs.

Open the workbenchSee harnesses and bases