| Benchmark | DeepSeek-V3.2 | GPT-5.5 |
|---|---|---|
| AA Agentic Index | 39.8 | 47.4 |
| AA Intelligence | 32.8 | 56.3 |
| AA-LCR | 70.7 | 79 |
| AA-Omniscience | -22.5 | 20.5 |
| aime_2026 | 95.1 | 100 |
| apexAgents | 14.5 | 37.7 |
| arc_agi_1 | 57 | 95 |
| ARC-AGI-2 | 4 | 85 |
| arena_text_factuality | 1433 | 1482 |
| Artificial Analysis Coding Index | 44.2 | 74.9 |
| browsecomp | 51.4 | 84.4 |
| coding_arena_elo | 1368 | 1457 |
| critpt | 2.9 | 27.1 |
| cybergym | 17.3 | 81.8 |
| frontiermath_tier_4 | 2.1 | 35.4 |
| gdpval | 18.8 | 49.5 |
| GPQA Diamond | 84 | 93.6 |
| HealthBench | 46.9 | 56.5 |
| HLE | 40.8 | 52.2 |
| HLE (with tools) | 40.8 | 52.2 |
| hmmt_feb_2026 | 84.1 | 98.5 |
| hmmt_nov_2025 | 90.2 | 96.5 |
| IFBench | 61 | 75.9 |
| MCP Atlas | 62.2 | 82.8 |
| OmniScience Accuracy | 33 | 58 |
| OmniScience Non-Hallucination | 17.3 | 12 |
| ResearchRubrics | 55.8 | 64 |
| scicode | 39 | 56.1 |
| simpleqa_verified | 27.5 | 63.1 |
| SWE-bench Pro | 15.6 | 58.6 |
| SWE-bench Verified | 73.1 | 80.6 |
| TauBench V3 - Banking | 18.8 | 39 |
| Terminal-Bench 2.0 | 46.4 | 82.7 |
| Terminal-Bench 2.1 | 46.8 | 84.3 |
| Terminal-Bench Hard | 35.6 | 60.6 |
| tool_decathlon | 35.2 | 55.6 |
| toolathlon | 35.2 | 55.6 |
| vectara_answer_rate | 92.6 | 100 |
| vectara_avg_summary_length | 62 | 129.6 |
| vectara_factual_consistency | 93.7 | 90.7 |
| vectara_hallucination_rate ↓ | 6.3 | 9.3 |
| τ²-Bench Telecom (AA run) | 90.6 | 93.9 |
| τ³-Bench | 69.2 | 31.3 |
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.