| Benchmark | DeepSeek-V4-Flash | Qwen3.7 Plus Preview |
|---|---|---|
| AA Agentic Index | 61.3 | 20.8 |
| AA Intelligence | 51.8 | 39.4 |
| AA-LCR | 74.3 | 69 |
| AA-Omniscience | -14.3 | 2.4 |
| Agents' Last Exam | 25.2 | 33.6 |
| Apex (Pass@1) | 33 | 22.7 |
| arena_elo | 1439 | 1458 |
| arena_text_factuality | 1434 | 1454 |
| Artificial Analysis Coding Index | 69.1 | 55.9 |
| critpt | 16.6 | 9.1 |
| gdpval | 52.9 | 22.2 |
| GPQA Diamond | 90.8 | 90.3 |
| HLE | 38.6 | 35.6 |
| hmmt_feb_2026 | 94.8 | 92.9 |
| IFBench | 79.2 | 79.1 |
| imo_answer_bench | 88.4 | 86 |
| MCP Atlas | 69 | 73.2 |
| MMLU-Pro | 86.4 | 88.5 |
| nl2repo | 54.2 | 41.1 |
| OmniScience Accuracy | 40.4 | 22.5 |
| OmniScience Non-Hallucination | 15.6 | 74.5 |
| scicode | 49.9 | 51.3 |
| SWE-bench Multilingual | 73.3 | 75.8 |
| SWE-bench Pro | 52.6 | 57.6 |
| SWE-bench Verified | 79 | 77.7 |
| TauBench V3 - Banking | 39.4 | 17.9 |
| Terminal-Bench 2.0 | 56.9 | 70.3 |
| Terminal-Bench 2.1 | 82.7 | 61 |
| Terminal-Bench Hard | 38.6 | 47 |
| τ²-Bench Telecom (AA run) | 95.6 | 93 |
| τ³-Bench | 22.9 | 17.9 |
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.