Create a routed chat completion
Send a standard Chat Completions request. model accepts a task alias,
a version pin such as alias@v2, or a direct registered-provider
reference such as openai/gpt-5.6-sol. Task requests are traced automatically.
Authorizations
W&B API key whose personal/default entity or teams own the selected task and providers.
Headers
Opt a non-streaming task-routed or direct provider/model request into exact-response caching. Cache identity includes the effective upstream request body and forwarded path/query parameters. Direct requests are additionally isolated by entity, provider configuration, model, passthrough user identity, and any forwarded W&B Inference project. readWrite reads and populates the cache, readOnly reads without populating it, and writeOnly populates without reading. Only successful responses up to 8 MiB are stored. Invalid values return HTTP 400. If op-cache is also supplied, both headers must select the same mode. Response caching is not supported for streaming requests.
readWrite, readOnly, writeOnly OpenPipe-compatible alias for wandb-cache-mode. It accepts the same modes, and true is equivalent to readWrite. If both headers are supplied, they must select the same mode.
readWrite, readOnly, writeOnly, true Body
Task alias, alias@vN, or {provider}/{model-id}.
111 <= x <= 90071992547409911 <= x <= 9007199254740991