Public Benefits Bench v1.1
Professional Work · 2026-06-09
Social-services benchmark testing whether models can help people navigate SNAP benefits, eligibility, and application workflows. Accuracy and reported uncertainty are criterion-level; the number of questions is not this metric's Bernoulli denominator. The exact criterion denominator is not disclosed.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 76.9 |
| Muse Wonder Full | 73.7 |
| Muse Wonder | 70.2 |
| Gemini 4 Argon | 69.8 |
| MiMo-V2.6-Pro | 68.9 |
| Hy4 Preview | 68.6 |
| GLM-5.3 | 68.5 |
| Muse Spark 1.2 | 68.5 |
| Kimi K3 | 68.2 |
| Opus 4.8 | 68.1 |
| MiMo-V2.6-Flash | 67.6 |
| Claude Sonnet 5.5 | 67.2 |
| Qwen3.8-Max | 67.1 |
| Grok 4.6 | 66.8 |
| GPT-5.6 Sol | 66.5 |
| Ember-1 | 66.2 |
| Sonnet 5 | 66.0 |
| Grok 4.7 | 65.6 |
| Gemini 3.8 Flash | 65.3 |
| DeepSeek V4.1 Flash | 64.3 |
| MiniMax M3 | 64.1 |
| Claude Haiku 5.5 | 63.2 |
| DeepSeek-V4-Pro | 62.9 |
| Sonnet 4.6 | 62.4 |
| GPT-5.6 Terra | 62.4 |
| GLM-5.1 | 61.8 |
| Muse Spark 1.1 | 61.3 |
| GPT-5.6 Luna | 61.2 |
| GPT-5.5 | 60.9 |
| Gemini 3.5 Flash | 59.5 |
| Inkling-Small | 59.4 |
| GPT-6.1 Sol | 59.3 |
| Inkling | 58.5 |
| Step 5 Preview | 58.1 |
| GPT-6 Luna | 57.6 |
| GPT-6 Sol | 56.6 |
| Kimi K2.6 | 56.6 |
| Gemini 3.6 Flash | 56.6 |
| Haiku 4.5 | 54.3 |
| Gemini 3.1 Pro Preview | 53.8 |