Troubleshooting
The problems people hit most, with the likely cause and the fix for each.
| Symptom | Likely cause | Fix |
|---|---|---|
401 on every request | The key is missing, mistyped, expired or revoked, or it is sent in the wrong header (x-api-key, or bearer in lower case) | Send Authorization: Bearer blz_..., with a capital B. BlazeInfer does not read the Anthropic-style x-api-key header. Check the key in the console |
429 while you believe you have credit | A short overload, or the account is out of credit | Check the balance in the console. If you have credit, retry with exponential backoff |
429 and you are out of credit | API keys need a credit balance | Add credits in the console |
404 "The requested model is not currently available." | A wrong or old model ID (often copied from an older guide), or a model that is not available right now | Use a full ID from console |
503 "Upstream provider error." | The model is temporarily unavailable | Retry, or use another model |
400 with a generic message | The model rejected the request, and BlazeInfer replaced the model's own message | Check parameters against the API reference and simplify the request. Quote the X-Request-Id in a support request |
| Stream stops with no error | A stream failed after the response started | Treat it as a failure, and retry. A stream that ends without a finish_reason is incomplete |
| Tool calls never come back | qwen/qwen3.8-27b returns no tool call unless you force one with tool_choice | Set tool_choice to "required" or a named function, or use google/gemma-4-31b-it |
| Every message in a coding-agent session fails with "The upstream provider rejected the request.", even "hi" | The agent read an image (for example a screenshot), and the model cannot read images. The image stays in the session history. OpenCode 2.x sends images by default | Set the model's image input to match its Vision column, then continue the session. See OpenCode, Kilo Code or Pi |
| A coding agent fails with "The upstream provider rejected the request." on a small model | The agent's system prompt and tools, plus the output tokens it asks for, do not fit in the model's context window | Set the context and output limits in the agent's config, or use a larger-context model such as google/gemma-4-31b-it |
| Reasoning text is missing or in an unexpected field | The field name differs between models (reasoning_content or reasoning) | Read both fields; see Reasoning models |
| A non-streaming request times out | The answer takes longer than 4 minutes | Stream the response. See Streaming |
Error messages and what they mean
If a request fails, match the exact message below to find the most likely cause.
| Message (exact text) | Status | What it means | Most likely causes | What to do |
|---|---|---|---|---|
api key required | 401 | BlazeInfer found no key | No Authorization header, the key sent as x-api-key, or bearer in lower case | Send Authorization: Bearer blz_... |
invalid api key | 401 | BlazeInfer does not accept the key | A mistyped, expired, revoked or blocked key | Check the key in the console, or create a new one |
The upstream provider rejected the request. | 400 (or another 4xx) | The model rejected the request | 1. The prompt plus max_tokens is larger than the model's context window.2. The body is not valid JSON, or the messages field is missing.3. qwen/qwen3.8-27b rejects a developer role message.4. tools or tool_choice sent to qwen/qwen3.5-9b, which does not accept tools.5. An image sent to a model that cannot read images (Vision: No on console), for example a screenshot read by a coding agent. | Send the smallest request that works (model plus one message), then add fields back one at a time. Lower max_tokens or shorten the prompt. Send the X-Request-Id to support |
Upstream provider error. | 5xx | The model failed to respond | A temporary fault, or the model is unavailable | Retry with backoff, or use another model |
The requested model is not currently available. | 404 | The model is not available | A wrong model ID, or a path that does not exist (for example /v1/embeddings) | Check the ID against console and the path against the API reference |
Upstream rate limit exceeded. Please retry. | 429 | The model is overloaded | High load | Retry with backoff |
Upstream authentication error. | 401 / 403 | BlazeInfer's own call to the model failed | A platform fault, not your key | Contact support with the X-Request-Id |
no models are currently available, automatic model routing is not available, no healthy pods available … (plain text) | 503 | No capacity for this model, or auto is turned off | An outage, or a wrong model | Retry later, or choose another model |
Because the original reason is replaced with a generic message, work from the smallest request that succeeds and add fields back one at a time.
Before you ask for help
Collect:
- The
X-Request-Idfrom the failing response. - The model ID, the endpoint and the time of the request.
- The full request (with the key removed) and the full response.
Then see Support & status.