Alibaba Cloud / Qwen Team

Qwen3-235B-A22B-Thinking-2507

Zero-eval
#1MMLU-Redux
#1WritingBench
#1MMLU-ProX
+11 more

by Alibaba Cloud / Qwen Team

+
+
+
+
About

Qwen3 235B Thinking was developed as a reasoning-enhanced variant, designed to incorporate extended thinking capabilities into the large-scale Qwen3 architecture. Built to combine deliberate analytical processing with mixture-of-experts efficiency, it serves tasks requiring both deep reasoning and computational practicality.

+
+
+
+
Pricing Range
Input (per 1M)$0.30 -$0.30
Output (per 1M)$3.00 -$3.00
Providers1
+
+
+
+
Timeline
AnnouncedJul 25, 2025
ReleasedJul 25, 2025
+
+
+
+
License & Family
License
Apache 2.0
Base ModelQwen3 235B A22B
Performance Overview
Performance metrics and category breakdown

Overall Performance

25 benchmarks
Average Score
69.2%
Best Score
93.8%
High Performers (80%+)
11

Performance Metrics

Max Context Window
387.1K
+
+
+
+
All Benchmark Results for Qwen3-235B-A22B-Thinking-2507
Complete list of benchmark scores with detailed information
MMLU-Redux
text
0.94
93.8%
Self-reported
AIME 2025
text
0.92
92.3%
Self-reported
WritingBench
text
0.88
88.3%
Self-reported
IFEval
text
0.88
87.8%
Self-reported
Creative Writing v3
text
0.86
86.1%
Self-reported
MMLU-Pro
text
0.84
84.4%
Self-reported
HMMT25
text
0.84
83.9%
Self-reported
GPQA
text
0.81
81.1%
Self-reported
MMLU-ProX
text
0.81
81.0%
Self-reported
Include
text
0.81
81.0%
Self-reported
Showing 1 to 10 of 25 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+