Compare models
Pick up to 4 models for a side-by-side spec sheet. Hover any dotted metric name for what the test measures. Best value in each row is highlighted.
Claude Sonnet 5Qwen3 235B A22B Instruct 2507
| Claude Sonnet 5 Anthropic | Qwen3 235B A22B Instruct 2507 Alibaba (Qwen) | |
|---|---|---|
| Pricing | ||
| Input $/1M | $2.00 | $0.09 |
| Output $/1M | $10.00 | $0.55 |
| Blended $/1M (3:1) | $4.00 | $0.205 |
| Cache read $/1M | $0.20 | — |
| Cache write $/1M | $2.50 | — |
| Quality indices | ||
| Intelligence Index | 53.4 | 18.2 |
| Coding Index | 71.5 | — |
| Math Index | — | 71.7 |
| Agentic Index | 46.7 | — |
| Speed & latency | ||
| Output speed (tok/s) | 86 | 74 |
| Time to first token | 93.73s | 1.12s |
| Time to first answer token | 93.73s | 1.12s |
| Benchmarks (Artificial Analysis) | ||
| MMLU-Pro | — | 82.8% |
| GPQA Diamond | 91.1% | 75.3% |
| Humanity's Last Exam | 39.6% | 10.6% |
| LiveCodeBench | — | 52.4% |
| SciCode | 53.6% | 36.0% |
| MATH-500 | — | 98.0% |
| AIME | — | 71.7% |
| AIME 2025 | — | 71.7% |
| IFBench | — | 46.1% |
| AA-LCR (Long Context Reasoning) | 70.7% | 31.2% |
| Terminal-Bench Hard | — | 15.2% |
| Terminal-Bench 2.1 | 80.5% | — |
| τ²-Bench (Telecom) | — | 33.3% |
| τ³-Bench Banking | 28.2% | — |
| Design Arena (Elo) | ||
| Website | 1,314 | 1,085 |
| Web apps | 1,305 | — |
| Full-stack | 1,292 | — |
| Android native | 1,255 | — |
| UI components | 1,317 | 1,009 |
| Data viz | 1,281 | 1,095 |
| SVG | 1,239 | — |
| 3D | 1,321 | 1,053 |
| Game dev | 1,345 | 1,010 |
| Agentic game dev | 1,241 | — |
| Godot game dev | 1,257 | — |
| HTML slides | 1,233 | — |
| Python→PPTX slides | 1,252 | — |
| Code categories | 1,312 | 1,072 |
| ASCII art | 1,230 | — |
| Specs | ||
| Context window | 1M | 262K |
| Max output tokens | 128K | 16K |
| Input modalities | text, image, file | text |
| Tokenizer | Claude | Qwen3 |