Chat

POST /v1/chat/completions

Creates a chat completion on a text model. OpenAI-compatible request and response; synchronous — the reply is in the response (or streamed, see below). Credits are reserved for the estimated prompt plus min(max_tokens, default_max_tokens) output tokens and settled to the actual usage when the call ends; nothing is charged when no tokens were produced.

Headers

Header	Required	Description
Authorization	yes	Bearer <API key>
Content-Type	yes	application/json
Idempotency-Key	no	Idempotency-Key is optional here. Reusing one returns `409 idempotency_key_reused`.

Body

Field	Type	Description
model	string	Required. A text model id from `GET /v1/models`, e.g. `"qwen3.8-max"`.
messages	array	Required. `{role: "system" \| "user" \| "assistant", content: string \| parts[]}`. Content parts (`text`, `image_url`) as on the OpenAI API; `image_url` only on models with `vision: true`.
max_tokens	integer	Reply cap. ≤ the model's `max_output_tokens`; default `default_max_tokens`. Reasoning tokens count toward it.
temperature	number	0–2. Default from the model's default_params.
top_p	number	0–1.
stop	string[]	Up to 4 stop sequences.
stream	boolean	Default `false`. `true` → `text/event-stream`, see Streaming.
enable_thinking	boolean	Extension. Reasoning before the answer, returned as `reasoning_content`. Only on models with `thinking: true`.
web_search	boolean	Extension. Lets the model consult the web. Only on models with `web_search: true`.
reasoning_effort	string	`none` / `minimal` turn thinking off, `low` … `xhigh` turn it on. Only on models with `thinking: true`.
tools	array	OpenAI function definitions. Only on models with `tools: true`; see Tool calling.
tool_choice	string \| object	`auto` · `none` · `required` · `{type: "function", function: {name}}`.
response_format	object	`{type: "json_object"}` or `{type: "json_schema", json_schema: {name, schema}}`; see Structured output.

Not supported — rejected with 400 unsupported_parameter, never silently ignored: n > 1, functions (legacy), logprobs.

curl -s -X POST https://api.genmux.tech/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-max","messages":[{"role":"user","content":"Explain idempotency keys in two sentences."}],"max_tokens":200}'

Response `200 OK`

{
  "id": "0d8b6a3e-2f1c-4a5b-9c1d-7e2f3a4b5c6d",
  "object": "chat.completion",
  "created": 1756411279,
  "model": "qwen3.8-max",
  "choices": [
    {"index": 0, "message": {"role": "assistant", "content": "An idempotency key is …"}, "finish_reason": "stop"}
  ],
  "usage": {"prompt_tokens": 31, "completion_tokens": 58, "total_tokens": 89, "cost": 0.004, "cost_source": "vendor"}
}

usage.cost is the credits charged (decimal, up to 3 places); cost_source is "vendor" when the upstream reported the token counts and "estimated" when they were counted locally.

Limits

Field	Limit
max_tokens	≤ the model's max_output_tokens; default default_max_tokens. A prompt that leaves no room for a reply is a 400 context_length_exceeded.
stop	Up to 4 strings.
temperature	0–2.
messages	Any number of system, user and assistant turns within the context window. image_url parts only on models with vision: true.

Streaming

With "stream": true the response is text/event-stream. Every data: line carries one chat.completion.chunk; the last chunk has an empty choices array and the usage block (with the credits charged), and the stream ends with data: [DONE].

curl -sN -X POST https://api.genmux.tech/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-max","messages":[{"role":"user","content":"Explain idempotency keys in two sentences."}],"stream":true}'

Response · text/event-stream

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{"role":"assistant","content":"An"},"finish_reason":null}]}

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{"content":" idempotency"},"finish_reason":null}]}

…

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[],"usage":{"prompt_tokens":31,"completion_tokens":58,"total_tokens":89,"cost":0.004,"cost_source":"estimated"}}

data: [DONE]

If the upstream fails mid-stream, an error frame is sent instead of the usage chunk and the stream still ends with [DONE]; the reservation is settled against what was actually generated, and GET /v1/generations/{id} shows the final cost.

Mid-stream failure

data: {"error": {"code": "upstream_error", "message": "the model provider closed the connection"}}

data: [DONE]

Closing the connection stops generation. The tokens generated before the abort are billed; nothing else is.

OpenAI SDK

The endpoint speaks the OpenAI chat completions dialect, so the official SDKs work unchanged: point base_url at https://api.genmux.tech/v1 and pass your key as api_key. Model ids are ours (see GET /v1/models), not the vendor's.

curl -s -X POST https://api.genmux.tech/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-max","messages":[{"role":"user","content":"Explain idempotency keys in two sentences."}],"max_tokens":200}'

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.genmux.tech/v1", api_key=os.environ["API_KEY"])

completion = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{"role": "user", "content": "Explain idempotency keys in two sentences."}],
    max_tokens=200,
)
print(completion.choices[0].message.content)
print(completion.usage.model_extra["cost"], "credits")

stream = client.chat.completions.create(model="qwen3.8-max", messages=[{"role": "user", "content": "Hi"}], stream=True)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")
    if chunk.usage:
        print("\n", chunk.usage.total_tokens, "tokens,", chunk.usage.model_extra["cost"], "credits")

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.genmux.tech/v1", apiKey: process.env.API_KEY });

const completion = await client.chat.completions.create({
  model: "qwen3.8-max",
  messages: [{ role: "user", content: "Explain idempotency keys in two sentences." }],
  max_tokens: 200,
});
console.log(completion.choices[0].message.content, completion.usage.cost);

const stream = await client.chat.completions.create({ model: "qwen3.8-max", messages: [{ role: "user", content: "Hi" }], stream: true });
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
  if (chunk.usage) console.log(`\n${chunk.usage.total_tokens} tokens, ${chunk.usage.cost} credits`);
}

Supported: model, messages (system/user/assistant; image_url parts on vision models), max_tokens, temperature, top_p, stop, stream, plus the extensions enable_thinking and web_search passed as extra body fields.

Also supported: tools / tool_choice on models with tools: true (see Tool calling), response_format json_object / json_schema (see Structured output), reasoning_effort. Rejected with 400 unsupported_parameter: n > 1, the legacy functions field, logprobs. Accepted and ignored: stream_options, parallel_tool_calls, seed, user, metadata, store, logit_bias and message name.

Read usage from the completion (sync) or from the last chunk (stream): cost is the credits charged.

Tool calling

OpenAI function calling on the models whose catalogue entry says tools: true. Send tools + tool_choice (auto · none · required · {type: "function", function: {name}}); the reply comes back with message.tool_calls and finish_reason tool_calls; return results as role: tool messages with the matching tool_call_id. Function calling is emulated by the api — the tools are rendered into the system prompt and the reply is parsed back — so the wire format is exactly OpenAI's while the models never see a native tools field.

Models with tools: true right now: doubao-seed-2-pro, doubao-seed-2-code, kimi-k3, claude-opus-5-5, claude-fable-5-1, gpt-6-astra, gpt-5.6-terra, qwen3.8-max. A tools request to any other model is 400 tools_unsupported.

Python · OpenAI SDK

from openai import OpenAI
client = OpenAI(base_url="https://api.genmux.tech/v1", api_key="nb_live_...")
tools = [{"type": "function", "function": {"name": "get_weather", "description": "Weather by city",
          "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}}}]
r = client.chat.completions.create(model="qwen3.8-max",
        messages=[{"role": "user", "content": "Weather in Paris?"}], tools=tools)
call = r.choices[0].message.tool_calls[0]        # call.function.name == "get_weather"
# run your function, then send the result back:
r2 = client.chat.completions.create(model="qwen3.8-max", tools=tools, messages=[
    {"role": "user", "content": "Weather in Paris?"},
    r.choices[0].message,
    {"role": "tool", "tool_call_id": call.id, "content": "18°C, cloudy"},
])
print(r2.choices[0].message.content)

Response · first call

{
  "id": "0d8b6a3e-…",
  "object": "chat.completion",
  "model": "qwen3.8-max",
  "choices": [
    {"index": 0, "finish_reason": "tool_calls", "message": {"role": "assistant", "content": null,
      "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "get_weather", "arguments": "{\"city\": \"Paris\"}"}}]}}
  ],
  "usage": {"prompt_tokens": 96, "completion_tokens": 21, "total_tokens": 117, "cost": 0.003, "cost_source": "vendor"}
}

With stream: true the call arrives as delta.tool_calls chunks — index, id and name first, then the arguments in fragments — followed by a chunk with finish_reason tool_calls and the usual usage chunk.

Streaming · text/event-stream

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{"role":"assistant","tool_calls":[{"index":0,"id":"call_1","type":"function","function":{"name":"get_weather","arguments":""}}]},"finish_reason":null}]}

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{\"city\": \"Paris\"}"}}]},"finish_reason":null}]}

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

data: {"id":"0d8b6a3e-…","object":"chat.completion.chunk","model":"qwen3.8-max","choices":[],"usage":{"prompt_tokens":96,"completion_tokens":21,"total_tokens":117,"cost":0.003,"cost_source":"estimated"}}

data: [DONE]

Limits

Field	Limit
tools	≤ 128 definitions, ≤ 64 KiB in total. Each definition costs prompt tokens: about 40 per tool plus its JSON schema.
tool_choice	auto (default), none, required or a named function. required and a named function are instructions the model usually follows; small models occasionally answer in prose instead.
parallel calls	Models usually emit one tool call per turn; parallel_tool_calls is accepted and ignored. Agents loop, so multi-step tasks complete.
strict	Accepted and ignored — there is no constrained decoding upstream.
response_format	Cannot be combined with tools in one request.
max_tokens	Values above max_output_tokens are clamped, not rejected. Default reply budget on the tool models is 8192 tokens.

Verified configs for OpenCode, Aider, Cline, Roo Code and Continue, and the measured pass rates, are on the coding agents page →

Structured output

Pass response_format with {"type": "json_object"} or {"type": "json_schema", "json_schema": {"name", "schema"}} ({name, schema}, JSON Schema 2020-12) and the model is instructed to answer with JSON only. Non-streamed replies are validated: on a failure the api retries once with the validation error fed back to the model; a second failure is 502 json_invalid and both attempts are billed. Streamed replies pass through unvalidated. response_format and tools cannot be combined.

curl -s -X POST https://api.genmux.tech/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-max",
       "messages":[{"role":"user","content":"Extract the city and the date: meet me in Paris on 3 March"}],
       "response_format":{"type":"json_schema","json_schema":{"name":"event",
         "schema":{"type":"object","properties":{"city":{"type":"string"},"date":{"type":"string"}},"required":["city","date"],"additionalProperties":false}}}}'

Response · content is the JSON document

{"id": "…", "object": "chat.completion", "model": "qwen3.8-max", "choices": [{"index": 0, "message": {"role": "assistant", "content": "{\"city\": \"Paris\", \"date\": \"2026-03-03\"}"}, "finish_reason": "stop"}], "usage": {…}}

POST /v1/chat/completions

Headers

Body

Response 200 OK

Limits

Streaming

OpenAI SDK

Tool calling

Limits

Structured output

Response `200 OK`