Alibaba Cloud / Qwen Team

Qwen3 235B A22B

Zero-eval
#1Arena Hard
#1BBH
#1CRUX-O
+4 more

by Alibaba Cloud / Qwen Team

+
+
+
+
About

Qwen3 235B was developed as a large-scale model with 235 billion total parameters using a mixture-of-experts architecture activating 22 billion parameters. Built to provide frontier capabilities with computational efficiency through sparse activation, it represents Qwen's advancement into very large-scale model development.

+
+
+
+
Pricing Range
Input (per 1M)$0.10 -$0.20
Output (per 1M)$0.10 -$0.80
Providers4
+
+
+
+
Timeline
AnnouncedApr 29, 2025
ReleasedApr 29, 2025
+
+
+
+
Specifications
Training Tokens36.0T
+
+
+
+
License & Family
License
Apache 2.0
Performance Overview
Performance metrics and category breakdown

Overall Performance

23 benchmarks
Average Score
76.2%
Best Score
95.6%
High Performers (80%+)
10

Performance Metrics

Max Context Window
256.0K
Avg Throughput
38.0 tok/s
Avg Latency
1ms
+
+
+
+
All Benchmark Results for Qwen3 235B A22B
Complete list of benchmark scores with detailed information
Arena Hard
text
0.96
95.6%
Self-reported
GSM8k
text
0.94
94.4%
Self-reported
BBH
text
0.89
88.9%
Self-reported
MMLU
text
0.88
87.8%
Self-reported
MMLU-Redux
text
0.87
87.4%
Self-reported
MMMLU
text
0.87
86.7%
Self-reported
AIME 2024
text
0.86
85.7%
Self-reported
MGSM
text
0.84
83.5%
Self-reported
AIME 2025
text
0.81
81.5%
Self-reported
MBPP
text
0.81
81.4%
Self-reported
Showing 1 to 10 of 23 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+