Harvey LAB-AA v1.1 (hallucination-gated)
Professional Work
Artificial Analysis and Harvey's LAB-AA v1.1 evaluates agents using the Stirrup harness on 120 revised private legal tasks across 24 practice areas. Its hallucination-gated all-pass rate averages the share of three judges finding every rubric criterion satisfied, assigning zero to any task with a material hallucination against its source documents. The judges are GPT-6 Sol, Grok 4.7, and Claude Opus 5.5. Scores are explicitly not comparable with v1.0.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Wonder Full | 13.1 |
| Muse Wonder | 10.0 |
| Grok 4.7 | 9.4 |
| Muse Spark 1.3 | 8.9 |
| GPT-6 Astra | 8.6 |
| GPT-6.1 Sol | 6.9 |
| Claude Fable 5.1 | 6.4 |
| Kimi K3 | 5.3 |
| Claude Opus 5.5 | 4.2 |
| Gemini 4 Argon | 3.9 |
| GPT-6 Sol | 3.6 |
| GPT-6 Luna | 3.3 |
| Mistral Large 4 Preview | 3.1 |
| Claude Haiku 5.5 | 2.8 |
| Claude Sonnet 5.5 | 2.8 |
| DeepSeek V4.1 Flash | 1.7 |
| GLM-5.3 Flash | 0.8 |
| GLM-5.3 | 0.3 |
| Gemini 3.1 Pro Preview | 0.0 |
| Gemini 3.5 Flash-Lite | 0.0 |
| Gemini 3.8 Flash | 0.0 |
| Inkling | 0.0 |
| Inkling-Small | 0.0 |
| MiniMax M3 | 0.0 |