Speka API — live system status

Speka does not rely on the upstream provider's model listing — a listed model can still fail on inference. Instead, every model in the catalog is probed on a schedule with a real request: a short chat completion, an embedding call, or a small image generation. The result (HTTP status, response time, and returned content) is what drives the status shown on this page, and models that fail repeatedly are removed from the public catalog automatically until they recover.

A 503 response means the service is temporarily unavailable. Implement exponential backoff — start at one second and double up to 30 seconds maximum — and retry the request. Check this status page for the current component health; most incidents resolve within minutes of detection. If a specific model shows as unavailable here, switch to another model with the same capability — your API key and base URL stay the same.

Chat completions accept a conversation thread and return a generated text response — used for dialogue, summarization, and instruction-following. Embeddings accept text and return a dense numeric vector used for semantic search, similarity ranking, and retrieval-augmented generation (RAG). The two serve different model types and are tracked separately on the status page.