| Benchmark | Claude Opus 4.8 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 49.4 | 27.5 |
| AA Intelligence | 57.3 | 37.7 |
| AA-LCR | 73 | 73.3 |
| AA-Omniscience | 28.8 | -20 |
| Agents' Last Exam | 27 | 27.3 |
| aime_2026 | 100 | 94.1 |
| Artificial Analysis Coding Index | 74.3 | 53.7 |
| charxiv_rq | 80.5 | 78.4 |
| critpt | 20.9 | 1.1 |
| DeepSWE 1.1 | 59 | 13.3 |
| gdpval | 54.2 | 32 |
| GPQA Diamond | 93.6 | 87.8 |
| HLE | 57.9 | 24 |
| hmmt_feb_2026 | 96.7 | 84.3 |
| hmmt_nov_2025 | 96.5 | 90.7 |
| IFBench | 62.2 | 69.1 |
| imo_answer_bench | 83.5 | 80.8 |
| JobBench | 48.4 | 21.8 |
| livebench_agentic_coding | 50.5 | 39.3 |
| livebench_coding | 81.8 | 71.8 |
| livebench_data_analysis | 66 | 70.4 |
| livebench_instruction_following | 72 | 53.2 |
| livebench_language | 79.7 | 63.3 |
| livebench_math | 94.3 | 79.9 |
| livebench_reasoning | 89.2 | 70.3 |
| mathvision | 86.7 | 85.1 |
| MMMU-Pro | 78.9 | 75.8 |
| nl2repo | 69.7 | 36.2 |
| OmniScience Accuracy | 48.8 | 19.6 |
| OmniScience Non-Hallucination | 60.7 | 50.7 |
| OSWorld-Verified | 83.4 | 63.9 |
| scicode | 53.5 | 39.8 |
| SWE-bench Multilingual | 84.4 | 71.3 |
| SWE-bench Pro | 69.2 | 53.5 |
| SWE-bench Verified | 88.6 | 77.2 |
| TauBench V3 - Banking | 34.2 | 16.7 |
| Terminal-Bench 2.0 | 74.6 | 59.3 |
| Terminal-Bench 2.1 | 85 | 60.7 |
| Terminal-Bench Hard | 58.3 | 34.8 |
| τ²-Bench Telecom (AA run) | 94.4 | 94.2 |
| τ³-Bench | 27.6 | 15.3 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.