Immigration Bench - All-Pass
Professional Work · 2026-10-28
Share of Immigration Bench's 262 questions on which every rubric item was met, without web search.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 26.3 |
| Muse Wonder Full | 25.6 |
| Claude Opus 5.5 | 24.4 |
| GPT-6.1 Sol | 22.5 |
| Grok 4.7 | 22.5 |
| Claude Sonnet 5.5 | 22.1 |
| GPT-6 Astra | 22.1 |
| Muse Wonder | 21.8 |
| Kimi K3 | 21.4 |
| Muse Spark 1.2 | 21.4 |
| GLM-5.3 | 21.0 |
| Gemini 4 Argon | 20.6 |
| MiMo-V2.6-Pro | 20.6 |
| DeepSeek V4.1 Flash | 19.5 |
| Mistral Large 4 Preview | 18.7 |
| Muse Spark 1.3 | 18.3 |
| Grok 4.6 | 17.6 |
| DeepSeek V4 Pro 0813 | 17.2 |
| Gemini 3.8 Flash | 16.4 |
| GPT-6 Luna | 16.4 |
| Claude Haiku 5.5 | 16.0 |
| Gemini 3.5 Flash-Lite | 14.1 |
| MiniMax M3 | 12.6 |