Vals AI Harvey Legal Agent Bench v1.1: leaderboard
HLAB-Vals v1.1: 120 private legal tasks, all-criteria task pass rates averaged over GPT-5.5 and Claude Sonnet 4.6 judges. Removes 17 unsolvable criteria, replaces one document, and updates harness, context compaction and finish handling. All models rerun; distinct from archived v1 and LAB-AA.
Metric: Task Resolution (%). Source: www.vals.ai. Saturation forecast: Not forecast. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Argon (High) | 27.08 | #7 (Gemini 4 Argon) |
| 2 | Muse Wonder (Max) | 23.75 | #14 (Muse Wonder) |
| 3 | Muse Wonder Full (Max) | 23.33 | #10 (Muse Wonder Full) |
| 4 | Grok 4.7 (xHigh) | 18.33 | #27 (Grok 4.7) |
| 5 | Mistral Large 4 (High) | 17.5 | #95 (Mistral Large 4) |
| 6 | GLM-5.3 (Max) | 14.58 | #39 (GLM-5.3) |
| 7 | DeepSeek V4.1 Flash (1017) (Max) | 14.17 | #20 (DeepSeek V4.1 Flash (1017)) |
| 8 | MiMo-V2.6-Pro | 13.75 | #42 |
| 9 | MiMo-V2.6-Flash | 13.33 | #71 |
| 10 | Kimi K3 (Max) | 12.92 | #31 (Kimi K3) |
| 11 | Gemini 3.8 Flash (High) | 12.08 | #23 (Gemini 3.8 Flash) |
| 12 | DeepSeek V4 Pro (0813) (Max) | 10.83 | #53 (DeepSeek V4 Pro (0813)) |
| 13 | GLM-5.3 Flash (Max) | 10.42 | #67 (GLM-5.3 Flash) |
| 14 | Claude Sonnet 5.5 (Max) | 10.42 | #11 (Claude Sonnet 5.5) |
| 15 | Claude Opus 5.5 (Max) | 9.17 | #4 (Claude Opus 5.5) |
No result here: #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-harvey-legal-agent-bench-v1-1 · How It Works · Data refreshed daily, snapshot 2026-10-21.