Arena-Hard-Auto (GPT-4-1106 judge) — Judge-model tag marks original Arena-Hard methodology version, distinct from generic entries
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | DeepSeek-R1 | DeepSeek | 92.3 | 5 | 2026-06-04 |
| 2 | OpenAI o1-mini | OpenAI | 92 | 4 | 2026-06-04 |
| 3 | DeepSeek-V3 | DeepSeek | 85.5 | 4 | 2026-06-04 |
| 4 | Claude 3.5 Sonnet | Anthropic | 85.2 | 4 | 2026-06-04 |
| 5 | GPT-4o | OpenAI | 80.4 | 4 | 2026-06-04 |
| 6 | R1-Zero | DeepSeek | 53.6 | 1 | 2026-05-03 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.