Is Cerebras Down?

If Prismix shows Cerebras as "Operational" but you're having issues:

• API key rate limit exceeded — Cerebras enforces per-key rate limits on both requests per minute and tokens per minute. Because Cerebras inference is extremely fast (2000+ tokens/sec), it is easy to exhaust token-per-minute limits even with a small number of requests. Implement token bucket logic in your application and check the x-ratelimit-remaining-tokens response header.
• Model unavailable during maintenance window — Cerebras occasionally takes specific model checkpoints offline for updates or capacity rebalancing. During this window, requests for that model return a 503. Check status.cerebras.ai for scheduled maintenance notices and fall back to an alternative model ID in your application.
• Streaming response cut off mid-generation — Cerebras's high-throughput streaming can overwhelm downstream HTTP clients that use small receive buffers. If your HTTP client or framework drops the connection before the stream completes, the generation appears truncated. Increase the client read timeout to at least 60 seconds and ensure your streaming consumer processes tokens as they arrive rather than buffering the full response.
• Context length exceeded (128k token limit) — Cerebras models support up to 128k context tokens. Sending prompts or conversation histories that exceed this limit returns a 400 with a context length error. Implement context pruning (summarize or drop older messages) to keep the total under the limit.
• 503 during peak load hours — Cerebras's wafer-scale chip infrastructure is not infinitely elastic. During peak demand periods (especially US business hours), the API may return 503 for new requests while existing requests complete. Implement exponential backoff with jitter: start at 1 second, cap at 30 seconds, and retry up to 5 times.
• SDK version incompatible with current API — Cerebras ships an OpenAI-compatible API but with version-specific extensions. If you pin an older version of the cerebras-cloud-sdk package, new model names or response fields may not be recognized. Run pip install --upgrade cerebras-cloud-sdk to get the latest client.