What shipped

kimi-k3-chat calls Venice's exact kimi-k3 model for $0.05. It accepts a prompt, an optional system message, temperature, and an output cap. Prompts stop at 8,000 characters and responses stop at 768 tokens.

kimi-k3-agent-plan costs $0.08. Give it a goal with optional context and constraints. It returns a JSON plan with assumptions, ordered steps, acceptance tests, risks, and the first action.

Both responses name the model and provider. When Venice supplies token usage, the chat response includes that too.

Why two endpoints

Some callers want direct model access. Others want a finished artifact they can hand to another agent without writing a planning prompt or parsing loose prose.

The limits are part of the product contract. They keep response time and upstream cost bounded while making the price clear before payment.