Chat completions
Send a conversation to a BlazeInfer model and read the reply.
Chat completions are the core of the API. You send a list of messages and get one assistant reply.
POST https://blaze.e2enetworks.net/v1/chat/completionsMessages and roles
Each message has a role and content. The usual roles are:
system— instructions that set the model's behaviour.user— input from your application.assistant— earlier replies (send these back to keep context).tool— the result of a tool call (see Tool calling).
qwen/qwen3.8-27b rejects the developer role, so use system for
instructions.
curl https://blaze.e2enetworks.net/v1/chat/completions \
-H "Authorization: Bearer $BLAZE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3.8-27b",
"messages": [
{"role": "system", "content": "You are a concise assistant."},
{"role": "user", "content": "Summarise HTTP caching in two sentences."}
],
"max_tokens": 256
}'Common parameters
| Parameter | Meaning |
|---|---|
model | A model ID or alias from console |
messages | The conversation so far |
max_tokens | Cap on the reply length, including any reasoning tokens. Set it explicitly |
temperature | Sampling randomness |
top_p | Nucleus sampling |
stream | Return Server-Sent Events instead of one body — see Streaming |
tools / tool_choice | Function calling — see Tool calling |
stop | Stop sequences. These can also match inside the reasoning text |
seed | Best-effort determinism |
Reasoning counts towards max_tokens. A reasoning model can return empty
content with finish_reason: "length" if reasoning uses up the budget — raise
max_tokens, or turn reasoning off (see
Reasoning models).
Parameter support depends on the model. If a model rejects a parameter, you get
400 with the generic message "The upstream provider rejected the request." (see
Errors & retries).
Response shape
The reply is a standard OpenAI chat completion: id, object, created,
model, choices[] (each with message and finish_reason) and a usage
block with prompt_tokens, completion_tokens and total_tokens. Responses
also carry an X-Request-Id header.
When they reason, qwen/qwen3.8-27b and qwen/qwen3.5-9b start content with
two newlines, so trim the text before you show it.