Public Benefits Bench v1 - Web Search
Professional Work · 2026-06-09
The Web Search split of Public Benefits Bench v1 allows web search without multi-turn interaction. It is kept under the archived v1 benchmark because v1.1 publishes a different leaderboard without these four condition splits.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 67.7 |
| Opus 4.8 | 58.9 |
| Gemini 3.5 Flash | 54.1 |
| DeepSeek-V4-Pro | 54.0 |
| GPT-5.5 | 53.9 |
| MiniMax M3 | 53.6 |
| Sonnet 4.6 | 52.4 |
| GLM-5.1 | 50.7 |
| Grok 4.3 | 44.0 |
| Kimi K2.6 | 43.6 |
| Gemini 3.1 Pro Preview | 41.8 |
| Haiku 4.5 | 36.1 |
| Grok 4.1 Fast | 33.1 |