Qwen3 Max Thinking

POST /v1/chat/completions

Reasoning model with adaptive tool use (search, memory, code interpreter) and test-time scaling for higher accuracy on complex tasks.

This model is deprecated and will be retired on 2026-07-08. After that date, requests to this model will fail. Migrate to a successor model before then.

At a glance

Field	Value
Model id	`qwen3-max-thinking`
Input modalities	Text
Output modalities	Text
Context window	256K
Weight precision	-
Max output tokens	65,536
Region	Singapore
Features	reasoning, code_interpreter, web_search, thinking
Native inference	No
New	No
Supported endpoints	`POST /v1/chat/completions`, `POST /v1/responses`, `POST /v1/messages`
Deprecation date	2026-07-08

Pricing

Charge	Spec	Rate
Input	per 1M prompt tokens	<=32K $1.08 (was $1.20); 32K-128K $2.16 (was $2.40); 128K-256K $2.70 (was $3.00)
Output	per 1M generated tokens	<=32K $5.52 (was $6.00); 32K-128K $11.04 (was $12.00); 128K-256K $13.80 (was $15.00)
Web Search	per request	$0.015

Example request

$ curl https://api.empiriolabs.ai/v1/chat/completions \
>   -H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \
>   -H 'Content-Type: application/json' \
>   -d '{"model": "qwen3-max-thinking", "messages": [{"role":"user","content":"Hello"}]}'

Parameters

Parameter	Type	Required	Default	Description
`temperature`	number	no	`0.7`	Sampling temperature. 0 = deterministic, 2 = maximum randomness. · Range: 0 – 2
`top_p`	number	no	`0.9`	Nucleus sampling probability mass. Lower = more focused. · Range: 0 – 1
`max_tokens`	number	no	`4096`	Maximum tokens in the response. · Range: 1 – 65536
`stop`	string	no	-	Up to 4 strings where the model will stop generating further tokens.
`enable_thinking`	boolean	no	true	Enable extended thinking mode. Slower but improves reasoning-heavy tasks.
`tool_web_search`	boolean	no	false	Allow the model to perform web searches when needed.
`web_search_mode`	enum	no	`"standard"`	standard = single search, thorough = multiple deeper searches. · Allowed: `standard`, `thorough`
`tool_code_interpreter`	boolean	no	true	Allow the model to execute Python code in a sandbox to compute / analyze data.
`tool_web_extractor`	boolean	no	true	Allow the model to fetch and read content from URLs it discovers.
`disable_formatting`	boolean	no	false	Skip the EmpirioLabs Markdown formatting (citation [N] rewriting + References block when web search / tools were used). The raw upstream answer with plain [N] citations is returned.

Notes

Web search modes: Standard (efficient) or Thorough (comprehensive, requires thinking).

Per-tool billing (usage.tool_usage)

When this model invokes tools (web search, code interpreter, etc.) inside a single request, the response carries a normalized usage.tool_usage map alongside the token counts. The example below shows the shape — exact field names, units, and which tools appear can vary slightly per provider:

1 "usage": {
2   "prompt_tokens": 123,
3   "completion_tokens": 456,
4   "cost_usd": 0.0042,
5   "tool_usage": {"web_search": 3, "code_interpreter": 1}
6 }

The tool counts are already factored into cost_usd — they are surfaced for transparency so you can audit per-tool billing. The field is omitted when no tools were invoked.

Machine-readable schema: GET https://api.empiriolabs.ai/v1/models/qwen3-max-thinking.