Qwen 3 Max Thinking
Qwen 3 Max Thinking is Alibaba Cloud's trillion-parameter reasoning model that autonomously deploys built-in search, memory, and code interpreter tools during inference, achieving a score of 49.8 on Humanity's Last Exam with search enabled.
import { streamText } from 'ai'
const result = streamText({ model: 'alibaba/qwen3-max-thinking', prompt: 'Why is the sky blue?'})Playground
Try out Qwen 3 Max Thinking by Alibaba Cloud. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.
Qwen 3 Max Thinking
Qwen 3 Max Thinking
Providers
Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.
P50 throughput on live AI Gateway traffic, in tokens per second (TPS). Visit the docs for more info.
P50 time to first token (TTFT) on live AI Gateway traffic, in milliseconds. View the docs for more info.
Direct request success rate on AI Gateway and per-provider. Visit the docs for more info.
More models by Alibaba Cloud
| Model |
|---|
| 1M | 1.0s | 357tps | $0.32/M | $1.28/M | Read:$0.08/M Write:$0.5/M | — | — | +3 |   | 06/02/2026 | |||
| 991K | 1.6s | 109tps | $1.25/M | $3.75/M | Read:$0.25/M Write:$1.56/M | — | — | +1 | | 05/21/2026 | |||
| 240K | 1.5s | 117tps | $1.30/M | $7.80/M | Read: $0.26/M Write: $1.63/M | — | — | +1 | | 04/20/2026 | |||
| 1M | 1.5s | 110tps | $0.50/M | $3.00/M | Read: $0.1/M Write: $0.63/M | — | — | +3 | | 04/02/2026 | |||
| 1M | 1.5s | 157tps | $0.10/M | $0.40/M | Read:$0.0/M Write:$0.13/M | — | — | +3 | | 02/24/2026 | |||
| 262K | 0.3s | 143tps | $0.09/M | $0.10/M | — | — |    | 04/28/2025 |
About Qwen 3 Max Thinking
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 When Choosing a Provider
- Configuration: Thinking-mode responses can be substantially longer than standard completions, factor in output token volume when estimating per-request cost across providers.
- Zero Data Retention: AI Gateway does not currently support Zero Data Retention for this model. See the documentation for models that support ZDR.
- Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
When to Use Qwen 3 Max Thinking
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 when
- 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
Conclusion
Qwen 3 Max Thinking is built for situations where getting the right answer matters more than getting it quickly: graduate-level problem solving, real-time research synthesis, and agentic pipelines that need a model capable of recognizing its own knowledge gaps. The combination of autonomous tool orchestration and transparent reasoning chains fits complex, open-ended inference tasks.