Across 6 shared benchmarks, Claude Sonnet 4 scores higher on 2 and Claude Sonnet 4-20250514 (no thinking) on 0, with 4 level. The widest gap is aider_polyglot, where Claude Sonnet 4 scores 70.7 against 56.4.
| Benchmark | Claude Sonnet 4 | Claude Sonnet 4-20250514 (no thinking) |
|---|---|---|
| aider_polyglot | 70.7 | 56.4 |
| arena_vision | 1207 | 1176 |
| vectara_answer_rate | 98.6 | 98.6 |
| vectara_avg_summary_length | 145.8 | 145.8 |
| vectara_factual_consistency | 89.7 | 89.7 |
| vectara_hallucination_rate ↓ | 10.3 | 10.3 |
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.