Alibaba Cloud / Qwen Team

Qwen3-235B-A22B-Instruct-2507

Zero-eval
#1ZebraLogic
#1MultiPL-E
#1Creative Writing v3
+9 more

by Alibaba Cloud / Qwen Team

+
+
+
+
About

Qwen3 235B Instruct was created as the instruction-tuned version of Qwen3 235B, designed to follow user instructions while leveraging the model's large-scale architecture. Built to provide advanced instruction-following with efficient mixture-of-experts design, it serves applications requiring both capability and practical deployment.

+
+
+
+
Pricing Range
Input (per 1M)$0.15 -$0.15
Output (per 1M)$0.80 -$0.80
Providers1
+
+
+
+
Timeline
AnnouncedJul 22, 2025
ReleasedJul 22, 2025
+
+
+
+
License & Family
License
Apache 2.0
Performance Overview
Performance metrics and category breakdown

Overall Performance

25 benchmarks
Average Score
72.1%
Best Score
95.0%
High Performers (80%+)
8

Performance Metrics

Max Context Window
147.5K
+
+
+
+
All Benchmark Results for Qwen3-235B-A22B-Instruct-2507
Complete list of benchmark scores with detailed information
ZebraLogic
text
0.95
95.0%
Self-reported
MMLU-Redux
text
0.93
93.1%
Self-reported
IFEval
text
0.89
88.7%
Self-reported
MultiPL-E
text
0.88
87.9%
Self-reported
Creative Writing v3
text
0.88
87.5%
Self-reported
WritingBench
text
0.85
85.2%
Self-reported
CSimpleQA
text
0.84
84.3%
Self-reported
MMLU-Pro
text
0.83
83.0%
Self-reported
Include
text
0.80
79.5%
Self-reported
MMLU-ProX
text
0.79
79.4%
Self-reported
Showing 1 to 10 of 25 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+