Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4 vs Qwen3 235B A22B Instruct 2507
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Sonnet 4 wins on 8/13 benchmarks
Claude Sonnet 4 wins 8 of 13 shared benchmarks. Leads in coding · reasoning · language.
Category leads
coding·Claude Sonnet 4reasoning·Claude Sonnet 4general·Qwen3 235B A22B Instruct 2507knowledge·Qwen3 235B A22B Instruct 2507math·Qwen3 235B A22B Instruct 2507language·Claude Sonnet 4
Hype vs Reality
Attention vs performance
Claude Sonnet 4
#167 by perf·#11 by attention
Qwen3 235B A22B Instruct 2507
#147 by perf·#1 by attention
Best value
Qwen3 235B A22B Instruct 2507
20.3x better value than Claude Sonnet 4
Claude Sonnet 4
5.1 pts/$
$9.00/M
Qwen3 235B A22B Instruct 2507
102.8 pts/$
$0.47/M
Vendor risk
Who is behind the model
Anthropic
$965.0B·Tier 1
Alibaba (Qwen)
$293.0B·Tier 1
Head to head
13 benchmarks · 2 models
Claude Sonnet 4Qwen3 235B A22B Instruct 2507
Aider polyglot
Claude Sonnet 4 leads by +1.7
Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework.
Claude Sonnet 4
61.3
Qwen3 235B A22B Instruct 2507
59.6
ARC-AGI
Claude Sonnet 4 leads by +29.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Sonnet 4
40.0
Qwen3 235B A22B Instruct 2507
11.0
ARC-AGI-2
Claude Sonnet 4 leads by +4.7
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Sonnet 4
5.9
Qwen3 235B A22B Instruct 2507
1.3
Dtbench
Qwen3 235B A22B Instruct 2507 leads by +2.2
Claude Sonnet 4
61.8
Qwen3 235B A22B Instruct 2507
64.0
Fiction.LiveBench
Qwen3 235B A22B Instruct 2507 leads by +6.0
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
Claude Sonnet 4
46.9
Qwen3 235B A22B Instruct 2507
52.9
Lmca
Claude Sonnet 4 leads by +6.4
Claude Sonnet 4
34.1
Qwen3 235B A22B Instruct 2507
27.7
OpenCompass · AIME2025
Qwen3 235B A22B Instruct 2507 leads by +0.8
Claude Sonnet 4
68.7
Qwen3 235B A22B Instruct 2507
69.5
OpenCompass · GPQA-Diamond
Qwen3 235B A22B Instruct 2507 leads by +0.9
Claude Sonnet 4
74.6
Qwen3 235B A22B Instruct 2507
75.5
OpenCompass · HLE
Qwen3 235B A22B Instruct 2507 leads by +3.6
Claude Sonnet 4
8.7
Qwen3 235B A22B Instruct 2507
12.3
OpenCompass · IFEval
Claude Sonnet 4
88.3
Qwen3 235B A22B Instruct 2507
88.3
OpenCompass · LiveCodeBenchV6
Claude Sonnet 4 leads by +4.5
Claude Sonnet 4
47.5
Qwen3 235B A22B Instruct 2507
43.0
OpenCompass · MMLU-Pro
Claude Sonnet 4 leads by +3.8
Claude Sonnet 4
83.0
Qwen3 235B A22B Instruct 2507
79.2
WeirdML
Claude Sonnet 4 leads by +7.4
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4
46.1
Qwen3 235B A22B Instruct 2507
38.7
Full benchmark table
| Benchmark | Claude Sonnet 4 | Qwen3 235B A22B Instruct 2507 |
|---|---|---|
Aider polyglot Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework. | 61.3 | 59.6 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 40.0 | 11.0 |
ARC-AGI-2 ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. | 5.9 | 1.3 |
Dtbench | 61.8 | 64.0 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 46.9 | 52.9 |
Lmca | 34.1 | 27.7 |
OpenCompass · AIME2025 | 68.7 | 69.5 |
OpenCompass · GPQA-Diamond | 74.6 | 75.5 |
OpenCompass · HLE | 8.7 | 12.3 |
OpenCompass · IFEval | 88.3 | 88.3 |
OpenCompass · LiveCodeBenchV6 | 47.5 | 43.0 |
OpenCompass · MMLU-Pro | 83.0 | 79.2 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 46.1 | 38.7 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $3.00 | $15.00 | 200K tokens (~100 books) | $60.00 | |
| $0.14 | $0.80 | 262K tokens (~131 books) | $3.05 |