Across 6 shared benchmarks, Claude Sonnet 4-20250514 (no thinking) scores higher on 1 and Gemini 2.5 Pro on 5. The widest gap is vectara_hallucination_rate, where Gemini 2.5 Pro scores 7 against 10.3.
| Benchmark | Claude Sonnet 4-20250514 (no thinking) | Gemini 2.5 Pro |
|---|---|---|
| aider_polyglot | 56.4 | 82.2 |
| arena_vision | 1176 | 1262 |
| vectara_answer_rate | 98.6 | 99.1 |
| vectara_avg_summary_length | 145.8 | 106.4 |
| vectara_factual_consistency | 89.7 | 93 |
| vectara_hallucination_rate ↓ | 10.3 | 7 |
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.