Models & Pricing

A model writes CodeWork’s answers and decides which tools to call. A provider is the service CodeWork reaches it through. You can use any model in the catalog from a provider you have credentials for, and switch per run, per project, or mid-conversation in your editor.

Connect a Provider

Subscriptions

Sign in once and use the models included in your plan.

Logins are stored in ~/.codework/aikit/auth.json and refreshed automatically when they expire.

API Keys

Set the provider’s environment variable and CodeWork picks it up. No sign-in step.

These are the common ones. Any other models.dev provider with tool-calling models works the same way: run codework models providers for the full list of IDs.

Your Own Models

Local models (Ollama, LM Studio, vLLM) and private gateways are added in settings. See Custom Models.

Choose a Model

Pass --provider and --model on a run:

export OPENROUTER_API_KEY="..."

codework run --provider openrouter --model deepseek/deepseek-v4.1-flash "Plan the database migration"

Or set a default for every run in your settings:

{
	"model": { "provider": "openrouter", "id": "deepseek/deepseek-v4.1-flash", "thinkingLevel": "medium" },
}

Without any settings, CodeWork uses openai/gpt-5.6-luna at high thinking, which needs OPENAI_API_KEY.

In an editor, use the model and Thinking Effort pickers instead. They list every model from the providers CodeWork has credentials for. See ACP.

Thinking Levels

Reasoning models let you trade speed and cost for depth. CodeWork uses one scale for every provider and maps it to each provider’s own setting:

off · minimal · low · medium · high · xhigh · max

Each model supports a subset; ask for one it doesn’t have and CodeWork uses the next level up. Pass --thinking <level> on a run, or set thinkingLevel in settings. The catalog lists the levels for each Codex model.

Pricing

CodeWork doesn’t charge for models. You pay your provider, at their rates or through your plan. Every run ends with a usage summary showing tokens and an estimated cost from the catalog’s per-million-token prices:

── usage ───────────────────────────────────────────────────────────────
model    openai-codex/gpt-5.6-luna
tokens   6,172 input · 1,377 output · 412 reasoning · 19,581 total
cache    12,032 read · 0 write
cost     $0.003127 · 3 turns

See Catalog & Prices for the price of each model and how the estimate is worked out.

Related