02·Compare
Side by side, same effort tier
Pick up to four models. Every column is evaluated at the tier you choose (resolved to the nearest tier each model supports), so the comparison is what you'd actually deploy.
+ GPT-5.5+ Claude Fable 5+ Claude Opus 4.8+ Claude Sonnet 5+ Gemini 3 Pro+ Grok 4.1− GLM-5.2− DeepSeek-V3.2+ Kimi K2 Thinking− Qwen3-235B-A22B− Llama 4 Maverick
| Metric | Qwen3-235B-A22B Medium | Llama 4 Maverick Standard | DeepSeek-V3.2 Medium | GLM-5.2 Medium |
|---|---|---|---|---|
| Intelligence Index | 66.7 | 52.4 | 69.5 | 69.9 |
| Cost / typical task | $0.0080 | $0.0008 | $0.0015 | $0.0072 |
| $/M in · out | $0.70 · $2.80 | $0.27 · $0.85 | $0.28 · $0.42 | $0.60 · $2.20 |
| Context window | 128K | 1M | 128K | 200K |
| Median speed | 70 tok/s | 160 tok/s | 60 tok/s | 95 tok/s |
| MMLU-Pro | 80.9 | 80.0 | 82.9 | 83.4 |
| GPQA Diamond | 74.3 | 70.0 | 76.1 | 74.8 |
| Humanity's Last Exam | 16.5 | 9.0 | 20.2 | 22.9 |
| AIME 2025 | 72.1 | 40.0 | 75.5 | 71.8 |
| SWE-bench Verified | 63.1 | 44.0 | 66.8 | 69.3 |
| LiveCodeBench | 70.5 | 48.0 | 73.3 | 72.9 |
| τ²-bench | 67.8 | 52.0 | 71.5 | 73.0 |
| IFEval | 85.7 | 82.0 | 86.7 | 88.2 |
Hover a benchmark name for what it measures — full plain-language guide on /benchmarks· preview data.