Create Run
Authorizations
Body
1glm-5.2, grok-4.5, kimi-k3, minimax-m3, claude-opus-4.7, claude-opus-4.8, claude-opus-5, claude-fable-5, claude-sonnet-5, gpt-5.5, gpt-5.6, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gemini-3.6-flash, gemini-3.5-flash, gemini-3.1-pro, gemini-3-flash Optional provider-native request parameters for the selected model, written with the provider's own field names and values and forwarded unchanged (e.g. {"reasoning": {"effort": "high"}} for OpenAI, {"thinking": {"type": "adaptive"}} for Anthropic, {"thinkingConfig": {"thinkingLevel": "high"}} for Google). Supported paths and values are per-model; an unsupported path, value, or a model that accepts no parameters at all is rejected with 422. Omitting this field applies the selected model's configured default parameters; passing {} opts out of that default and leaves the provider's own defaults in place.
Browser config, owned by the browser row (never the run). Follow-ups: omitted fields inherit the session's prior browser; explicit values apply only to newly provisioned browsers (a live browser is reused as-is). Profile is the exception: a different profile → 400; an explicit value mismatching the live browser's profile (incl. null vs a profiled browser) declines reuse and provisions fresh. A live browser already matching the requested profile is reused.
20Opt-in LLM judgement of the finished run. Presence of this object turns the judge on (send {} for defaults); CP judges the run's trajectory after it goes terminal and surfaces the verdict as RunSummary.judgement (null until it lands). The judge's LLM call is billed to the run.
x > 0