# Qwen 3 Max Thinking

Qwen 3 Max Thinking adds autonomous tool orchestration and extended step-by-step reasoning to Alibaba Cloud's trillion-parameter Qwen3-Max line, selecting and invoking built-in search, memory, and code interpreter capabilities without requiring explicit user direction.

- **Model ID:** `alibaba/qwen3-max-thinking`
- **Type:** chat
- **Providers:** alibaba
- **Context window:** 256,000
- **Maximum output tokens:** 65,536
- **Pricing:** $1.2/1M input tokens, $6/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/qwen3-max-thinking

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'alibaba/qwen3-max-thinking',
  prompt: 'Why is the sky blue?'
})
```

## About

Qwen 3 Max Thinking, released on January 23, 2026, extends the Qwen3-Max architecture with a dedicated extended reasoning mode and integrated autonomous tool use. When Qwen 3 Max Thinking encounters a question that exceeds its internal knowledge or requires computation, it independently decides whether to trigger its Search tool (for current information), Memory tool (for cross-turn context persistence), or Code Interpreter (for numerical verification and data processing), without you needing to specify which tool applies.

This autonomous tool selection is a meaningful architectural distinction. Rather than exposing tool invocation as an explicit user-facing control, Qwen 3 Max Thinking treats it as an internal reasoning step, making the interaction feel more like working with a capable assistant that knows when to check its work. The design is intended to reduce hallucination risk on factual queries by defaulting to retrieval when confidence is low, and to improve numerical accuracy by routing computations through an interpreter.

Qwen 3 Max Thinking's thinking mode exposes its reasoning chain before delivering a final answer, providing transparency into multi-step problem decomposition. On Humanity's Last Exam, a benchmark of approximately 3,000 graduate-level questions spanning mathematics, science, and engineering, Qwen 3 Max Thinking with search enabled scored 49.8, competitive with other models on the same benchmark in Alibaba Cloud's published comparisons. You can access Qwen 3 Max Thinking through AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python.

## What to consider

Thinking-mode responses can be substantially longer than standard completions, factor in output token volume when estimating per-request cost across providers.

## When to use

### Best For

- Hard scientific, mathematical, or engineering questions that benefit from visible chain-of-thought reasoning
- Research assistance workflows where factual grounding via real-time search reduces the risk of stale or invented information
- Automated agents that need to self-correct numerical outputs by routing calculations through a code interpreter
- Multi-turn conversations that require the model to remember and adapt to user preferences over an extended session
- Complex task pipelines where the model must autonomously decide which tool to invoke without explicit user prompting

### Consider Alternatives When

- Token budget is tightly constrained, since thinking-mode responses consume substantially more output tokens than direct-answer models
- Your task requires fast, low-latency completions and extended reasoning is unnecessary overhead
- You need precise manual control over which tools fire and when, since autonomous tool selection may not align with strictly orchestrated pipelines
- Creative or conversational tasks where step-by-step reasoning adds friction rather than value

## Best for

- **STEM and engineering questions:** Hard scientific, mathematical, or engineering problems that benefit from visible chain-of-thought reasoning
- **Search-grounded research assistance:** Workflows where real-time search reduces the risk of stale or invented information
- **Self-correcting numerical agents:** Automated agents that route calculations through a code interpreter to verify outputs
- **Multi-turn personalized conversations:** Extended sessions where the model must remember and adapt to user preferences
- **Autonomous tool orchestration:** Complex pipelines where the model decides which tool to invoke without explicit user prompting

## Consider alternatives

- **Tight token budgets:** Thinking-mode responses consume substantially more output tokens than direct-answer models
- **Low-latency requirements:** Fast completions matter more than extended reasoning, which becomes unnecessary overhead
- **Deterministic tool control:** Autonomous tool selection may not align with strictly orchestrated pipelines that need manual control
- **Creative or conversational tasks:** Step-by-step reasoning adds friction rather than value in these contexts

## Frequently asked questions

### What tools does Qwen 3 Max Thinking invoke autonomously?

The model has three integrated tools: Search (for retrieving current information), Memory (for tracking user preferences and context across turns), and Code Interpreter (for executing code to verify or compute results). It selects among them without user prompting.

### How does the thinking mode differ from standard completion mode?

In thinking mode, the model generates an extended internal reasoning trace before producing its final answer. This trace is visible in the response and reflects multi-step problem decomposition, which improves accuracy on hard reasoning tasks at the cost of more output tokens.

### What did the model score on Humanity's Last Exam?

With its Search tool enabled, Qwen 3 Max Thinking scored 49.8 on Humanity's Last Exam, a benchmark of ~3,000 graduate-level questions across math, science, and engineering.

### Which SDKs are compatible with this model via AI Gateway?

You can access Qwen 3 Max Thinking through AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python.

### Does autonomous tool use mean I can't control when tools fire?

You can't manually control which tools fire. The model selects tools as an internal reasoning step. If your application requires deterministic, manually orchestrated tool calls, the non-thinking Qwen3-Max variant with explicit function-calling schemas is a better fit.

### How does the Search tool reduce hallucinations?

When Qwen 3 Max Thinking's internal confidence on a factual claim is low, it automatically triggers a search query to retrieve current information rather than generating an answer from its internal knowledge alone, which reduces the frequency of confidently stated incorrect facts.

### What is the token cost model for thinking-mode responses?

Thinking-mode completions generate reasoning traces that can be substantially longer than direct answers. Output tokens from both the reasoning chain and the final response are billed.

## Links

- [Model page](https://vercel.com/ai-gateway/models/qwen3-max-thinking)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://modelstudio.console.alibabacloud.com/)
- [Provider pricing](https://www.alibabacloud.com/help/en/model-studio/model-pricing)