SnorkelUnderwrite 2.0
Professional Work
Pass@1 on 200 commercial property-and-casualty underwriting tasks with partial information, cross-source evidence, role permissions and stateful workflows. Experts review scenarios, tools, rubrics and verifiers. Native effort and agent scaffold are not disclosed.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.7 | 30.5 |
| GLM-5.3 | 30.4 |
| Kimi K3 | 27.9 |
| Claude Fable 5.1 | 24.0 |
| Claude Opus 5.5 | 22.7 |
| GPT-6 Astra | 21.2 |
| Muse Spark 1.3 | 19.6 |
| Gemini 3.8 Flash | 19.1 |
| Claude Opus 5 | 18.3 |
| Grok 4.6 | 17.8 |