Legal Agent Benchmark
Agentic legal document reasoning and multi-step legal problem-solving benchmark.
Models ranked
6
tracked on this benchmark
Score band
20.0 – 2.1
best → lowest tracked
Snapshot trend
+10.77
Jun 9 → Oct 17 · 3 models
Leaderboard
Tracked models ranked by Accuracy (higher is better).
Configuration: Vals AI default provider Meta; max output 256,000, reasoning effort xhigh
Notes: Vals AI leaderboard (https://www.vals.ai/benchmarks/hlab), page 'Updated 10/16/2026', retrieved 2026-10-17. Vals model page https://www.vals.ai/models/meta_muse-spark-1.1 shows the hyperparameters and 20.00% ±2.94 (rank 5/80). Metric is the task all-pass rate: a task passes only if 100% of its criteria pass, averaged over two LLM judges (GPT 5.5 medium reasoning, Claude Sonnet 4.6). Vals AI is the sole source for this row (not a lab-restatement figure).
Configuration: Vals AI default provider Google; temperature 1, max output 262,144, reasoning effort high
Notes: Vals AI leaderboard (https://www.vals.ai/benchmarks/hlab), page 'Updated 10/16/2026', retrieved 2026-10-17. Vals model page https://www.vals.ai/models/google_gemini-4-argon shows the hyperparameters. Metric is the task all-pass rate: a task passes only if 100% of its criteria pass, averaged over two LLM judges (GPT 5.5 medium reasoning, Claude Sonnet 4.6). Vals model page: ±3.31. The lab's restatement (https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) reads 19.6; it matches after rounding and is not the source of this row.
Notes: z.ai blog (2026-10-12): 19.2% all-pass rate on Harvey LAB; independently evaluated by Vals (https://www.vals.ai/models/zai_glm-legal).
Configuration: Vals AI default provider Mistral AI; temperature 1, top-p 0.95, max output 256,000, reasoning effort high
Notes: Vals AI leaderboard (https://www.vals.ai/benchmarks/hlab), page 'Updated 10/16/2026', retrieved 2026-10-17. Vals model page https://www.vals.ai/models/mistralai_mistral-large-4 shows the hyperparameters. Metric is the task all-pass rate: a task passes only if 100% of its criteria pass, averaged over two LLM judges (GPT 5.5 medium reasoning, Claude Sonnet 4.6). Vals model page: ±2.96. The lab's restatement (https://mistral.ai/news/mistral-large-4/) reads 15.8 (chart read, "Vals.ai Harvey LAB"); it matches after rounding and is not the source of this row.
Notes: Unstarred in Anthropic's shared Fable 5/Mythos 5 launch table, so this is a Claude Fable 5 score.
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: Jun 15, 2026