Harvey LAB-AA
Legal work across 120 tasks and 24 practice areas, scored against task-specific grading criteria.
Legal reasoningArtificial Analysis42 models run by Artificial AnalysisUpdated 8 October 2026
How it is run
Artificial Analysis evaluates the Harvey Legal Agent Benchmark with criterion-level rubric judging and reports its own independent score.
What it cannot tell you
These AA runs are separate from Vals AI’s HLAB implementation. Grading criteria, judge models and agent settings affect the result.
- Leader
- Muse Spark 1.395.48% · Artificial Analysis results, grouped here by release
- Top-ten spread
- 4.40%Leader minus tenth place, or last place in smaller fields.
- Median
- 85.09%of the 42 rows published
- No published result
- 26of 46 models on this site
Artificial Analysis results by release
Highest published effort per model release, grouped here from AA’s public results. Grouping ranks are derived here; source settings and model-publisher reports stay separate.
| # | Model | Score | Effort |
|---|---|---|---|
| 1 | Muse Spark 1.3 | 95.48% | xhigh |
| 2 | Kimi K3 | 94.64% | max |
| 3 | Qwen3.8 Max (0902) | 93.65% | Not specified |
| 4 | Claude Fable 5 | 93.56% | max |
| 5 | Claude Opus 5 | 93.46% | max |
| 6 | Claude Sonnet 5.5 | 93.07% | max |
| 7 | Claude Fable 5.1 | 93.02% | max |
| 8 | Grok 4.5 | 92.42% | high |
| 9 | Claude Opus 5.5 | 91.20% | max |
| 10 | Claude Opus 4.8 | 91.08% | max |
| 11 | GLM-5.2 | 90.97% | max |
| 12 | Gemini 3.7 Flash | 90.66% | high |
| 13 | Claude Sonnet 5 | 90.07% | max |
| 14 | Claude Haiku 5.5 | 89.87% | max |
| 15 | MiniMax-M3 | 88.41% | Not specified |
| 16 | GPT-5.6 Luna | 87.90% | max |
| 17 | GPT-5.6 Sol | 87.18% | max |
| 18 | GPT-5.5 | 86.28% | xhigh |
| 19 | Claude Sonnet 4.6 | 85.97% | max |
| 20 | GPT-5.6 Terra | 85.18% | max |
| 21 | Gemini 3.6 Flash | 85.15% | high |
| 22 | Kimi K2.7 Code | 85.02% | Not specified |
| 23 | DeepSeek V4 Pro 0424 | 84.40% | max |
| 24 | Kimi K2.6 | 84.12% | Not specified |
| 25 | Qwen3.7 Max | 83.43% | Not specified |
| 26 | Qwen3.6 27B | 82.30% | Not specified |
| 27 | Gemini 3.5 Flash | 82.08% | high |
| 28 | Qwen3.7 Plus | 81.79% | Not specified |
| 29 | Nemotron 3 Ultra 550B A55B | 81.72% | Not specified |
| 30 | DeepSeek V4 Flash 0420 | 81.33% | max |
| 31 | MiMo-V2.5-Pro | 73.25% | Not specified |
| 32 | Step 3.7 Flash | 72.73% | high |
| 33 | Qwen3.5 397B A17B | 72.38% | Not specified |
| 34 | Mistral Medium 3.5 | 69.10% | high |
| 35 | Grok 4.3 | 68.39% | high |
| 36 | Claude 4.5 Haiku | 61.05% | Not specified |
| 37 | GPT-5.4 mini | 60.75% | xhigh |
| 38 | Gemini 3.1 Pro Preview | 58.85% | Not specified |
| 39 | GPT-5.4 nano | 52.24% | xhigh |
| 40 | Gemma 4 31B | 47.23% | Not specified |
| 41 | Gemini 3.1 Flash-Lite | 31.15% | high |
| 42 | gpt-oss-120b | 13.89% | high |
No published result tracked
26 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.
Step 5 PreviewGPT-6 LunaGLM 5.3 FlashQwen3.8 2.4T A95BQwen3.8-Flash-NextMiMo-V2.6-FlashGrok 4.6Claude Opus 4.7Muse Spark 1.2GPT-5.4DeepSeek V4.1 Flash 1017GLM-LegalBeamGemini 4 ArgonGPT-6 AstraGPT-6 SolGPT-6.1 SolGemini 3.8 FlashGrok 4.7DeepSeek V4.1 FlashDeepSeek V4 ProQwen3.8 MaxGLM 5.3Hunyuan Hy4 PreviewMiMo V2.6 ProInkling
Sources
Every page used for this benchmark, read on 17 October 2026.
- Benchmark site or the page that publishes itartificialanalysis.ai/evaluations/harvey-lab-aa
- Independent resultsartificialanalysis.ai/evaluations/harvey-lab-aa