Qwen3-Max-Thinking

Released

Qwen3-Max-Thinking is an AI model released by Qwen on Monday, Jan 26 2026, 124 days after Qwen3-Max. It is a >1T parameterThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. model with a 256k token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover SWE-Bench Verified, LiveCodeBench, Humanity's Last Exam, GPQA Diamond, IFBench, and MMLU-Pro.

Benchmarks

Coding

SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better.
75.3%
#33 of 95
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.
85.9%
#12 of 86

Reasoning & science

Humanity's Last ExamMultidisciplinary reasoning — accuracy on the 2,158 text-only questions without tools
30.2%text only, no tools
49.8%with tools
text only, no tools
#9 of 27
with tools
#37 of 55
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
87.4%
#41 of 145
IFBenchInstruction following — Tests whether the AI can follow detailed instructions and satisfy multiple constraints at once. Higher is better.
70.9%
#10 of 16

Knowledge work

MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark
85.7%
#6 of 58

Compare Qwen3-Max-Thinking with

Qwen3-Max-Thinking

About

Qwen3-Max-Thinking was announced on January 26, 2026 as the reasoning revision of Alibaba's Qwen3 Max tier. It could invoke retrieval or a code interpreter when needed. Alibaba also tested a higher-compute configuration that reused insights across successive reasoning attempts.

Qwen's standard launch table reported 75.3 on SWE-bench Verified. On Humanity's Last Exam, it scored 30.2 on the text-only subset without tools and 49.8 in the tool-assisted evaluation. Alibaba served the release under the dated qwen3-max-2026-01-23 identifier.

Frequently asked questions

Qwen3-Max-Thinking was released by Qwen on Monday, Jan 26 2026.

All Qwen releases

42 tracked