Vals Harvey's Legal Agent Benchmark (Harvey's Legal Agent Benchmark)
We show this table for reference; we do not rank on it.
Tests an agent's ability to complete legal work using documents, spreadsheets, presentations, and file-system tools
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Harvey's Legal Agent Benchmark score on Harvey's Legal Agent Benchmark — October 12, 2026
We mirror the published harvey's legal agent benchmark score view for Harvey's Legal Agent Benchmark. Muse Spark 1.2 leads the public snapshot at 25.42%, followed by Muse Spark 1.3 Max (23.75%) and Muse Spark 1.3 (22.92%). We do not use these results to rank models overall.
Muse Spark 1.2
Meta
xhigh reasoning
Muse Spark 1.3 Max
Meta
max reasoning
Muse Spark 1.3
Meta
xhigh reasoning
79 modelsAgenticCurrentDisplay onlyUpdated October 12, 2026
Harvey's Legal Agent Benchmark score table (79 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Spark 1.2Metaxhigh reasoning | 25.42% | Not reported | Closed |
| 2 | Muse Spark 1.3 MaxMetamax reasoning | 23.75% | Not reported | Unknown |
| 3 | Muse Spark 1.3Metaxhigh reasoning | 22.92% | Not reported | Closed |
| 4 | Muse Wonder FullMetamax reasoning | 21.67% | Not reported | Closed |
| 5 | Muse Spark 1.1Metaxhigh reasoning | 20.00% | Not reported | Closed |
| 6 | Gemini 4 ArgonGooglehigh reasoning | 19.58% | Not reported | Closed |
| 7 | GLM LegalZhipu AImax reasoning | 19.17% | Not reported | Unknown |
| 8 | Muse WonderMetamax reasoning | 17.08% | Not reported | Closed |
| 9 | Mistral Large 4Mistralhigh reasoning | 15.83% | Not reported | Pending |
| 10 | Grok 4.6xAIhigh reasoning | 15.83% | Not reported | Closed |
| 11 | Grok 4.5xAIhigh reasoning | 12.92% | Not reported | Closed |
| 12 | Kimi K3Moonshot AImax reasoning | 12.92% | Not reported | Pending |
| 13 | Grok 4.7xAIxhigh reasoning | 12.50% | Not reported | Closed |
| 14 | MiMo-V2.6-FlashXiaomi | 11.25% | Not reported | Open |
| 15 | Qwen3.8-27BAlibabaxhigh reasoning | 11.25% | Not reported | Open |
| 16 | Claude Fable 5Anthropic | 11.25% | Not reported | Closed |
| 17 | MiMo-V2.6-ProXiaomi | 10.83% | Not reported | Open |
| 18 | Qwen3.8 MaxAlibaba | 10.42% | Not reported | Open |
| 19 | Gemini 3.8 FlashGooglehigh reasoning | 10.00% | Not reported | Closed |
| 20 | Claude Opus 4.8Anthropic | 9.58% | Not reported | Closed |
| 21 | Step 5 PreviewStepFun | 9.17% | Not reported | Open |
| 22 | Hy4 previewTencent | 9.17% | Not reported | Open |
| 23 | Ember-1Fireworks | 9.17% | Not reported | Closed |
| 24 | Gemini 3.7 FlashGooglehigh reasoning | 8.75% | Not reported | Closed |
| 25 | DeepSeek V4 Flash 0731DeepSeekhigh reasoning | 8.33% | Not reported | Open |
| 26 | GLM-5.3Z.AImax reasoning | 8.33% | Not reported | Open |
| 27 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 7.50% | Not reported | Open |
| 28 | GLM-5.2Z.AImax reasoning | 7.08% | Not reported | Open |
| 29 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 6.67% | Not reported | Open |
| 30 | Claude Opus 4.7Anthropic | 6.67% | Not reported | Closed |
| 31 | GLM-5.3-FlashZ.AImax reasoning | 6.67% | Not reported | Open |
| 32 | Claude Opus 5Anthropic | 6.67% | Not reported | Closed |
| 33 | Claude Fable 5.1Anthropic | 6.67% | Not reported | Closed |
| 34 | GPT-6 AstraOpenAImax reasoning | 5.42% | Not reported | Closed |
| 35 | GPT-6.1 SolOpenAImax reasoning | 5.42% | Not reported | Closed |
| 36 | Claude Sonnet 4.6Anthropic | 5.00% | Not reported | Closed |
| 37 | Claude Sonnet 5Anthropic | 5.00% | Not reported | Closed |
| 38 | MiniMax M3MiniMax | 4.17% | Not reported | Open |
| 39 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 3.75% | Not reported | Open |
| 40 | GPT-5.5OpenAIxhigh reasoning | 3.75% | Not reported | Closed |
| 41 | Claude Opus 5.5Anthropic | 3.75% | Not reported | Closed |
| 42 | Gemini 3.6 FlashGooglehigh reasoning | 3.33% | Not reported | Closed |
| 43 | GPT-6 LunaOpenAImax reasoning | 2.92% | Not reported | Closed |
| 44 | Claude Sonnet 5.5Anthropic | 2.92% | Not reported | Closed |
| 45 | Gemini 3.5 FlashGooglehigh reasoning | 2.50% | Not reported | Closed |
| 46 | GPT-5.6 SolOpenAImax reasoning | 2.50% | Not reported | Closed |
| 47 | MiMo-V2.5-ProXiaomi | 2.08% | Not reported | Closed |
| 48 | InklingThinking Machines Lab0.99 reasoning | 2.08% | Not reported | Open |
| 49 | MiMo-V2.5Xiaomi | 1.67% | Not reported | Closed |
| 50 | GPT-6 SolOpenAImax reasoning | 1.67% | Not reported | Closed |
| 51 | Kimi K2.6Moonshot AI | 1.67% | Not reported | Open |
| 52 | Inkling-SmallThinking Machines Lab0.99 reasoning | 1.67% | Not reported | Open |
| 53 | Qwen3.7 MaxAlibaba | 1.67% | Not reported | Closed |
| 54 | Ling 3.0 Flash 2607Ant | 1.25% | Not reported | Unknown |
| 55 | GPT-5.6 LunaOpenAImax reasoning | 1.25% | Not reported | Closed |
| 56 | Qwen3.6 PlusAlibaba | 1.25% | Not reported | Closed |
| 57 | Claude Haiku 5.5Anthropic | 1.25% | Not reported | Closed |
| 58 | Claude Haiku 4.5 ThinkingAnthropic | 0.83% | Not reported | Closed |
| 59 | GPT-5.6 TerraOpenAImax reasoning | 0.83% | Not reported | Closed |
| 60 | Grok 4.3xAIhigh reasoning | 0.42% | Not reported | Closed |
| 61 | Nemotron 3 Ultra 550b A55bNvidia | 0.42% | Not reported | Unknown |
| 62 | Mistral Medium 3.5Mistral AIhigh reasoning | 0.42% | Not reported | Unknown |
| 63 | Gemini 3.1 Flash Lite PreviewGooglehigh reasoning | 0.00% | Not reported | Unknown |
| 64 | Mercury 2.5Inceptionhigh reasoning | 0.00% | Not reported | Closed |
| 65 | Gemini 3 Flash PreviewGooglehigh reasoning | 0.00% | Not reported | Unknown |
| 66 | Gemini 3.5 Flash-LiteGooglehigh reasoning | 0.00% | Not reported | Closed |
| 67 | Ling 3.0 Flash Af Rc3Ant | 0.00% | Not reported | Unknown |
| 68 | GPT-5.4 nanoOpenAIhigh reasoning | 0.00% | Not reported | Closed |
| 69 | MiniMax M2.7MiniMax | 0.00% | Not reported | Open |
| 70 | Gemini 3.1 Pro PreviewGooglehigh reasoning | 0.00% | Not reported | Unknown |
| 71 | Laguna M.1Poolside | 0.00% | Not reported | Closed |
| 72 | GLM-5.1Z.AI | 0.00% | Not reported | Open |
| 73 | Grok 4.20 0309 ReasoningxAI | 0.00% | Not reported | Unknown |
| 74 | Qwen3.7 PlusAlibaba | 0.00% | Not reported | Closed |
| 75 | GPT-5.4 miniOpenAIxhigh reasoning | 0.00% | Not reported | Closed |
| 76 | GPT-5.4OpenAIxhigh reasoning | 0.00% | Not reported | Closed |
| 77 | Laguna XS.2Poolside | 0.00% | Not reported | Open |
| 78 | Kimi K2.5 ThinkingMoonshot AI | 0.00% | Not reported | Unknown |
| 79 | Nemotron Lightning 3p5 30b A3bFireworks AI | 0.00% | Not reported | Unknown |
How Harvey's Legal Agent Benchmark is shown here
BenchLM mirrors the public Vals AI Harvey's Legal Agent Benchmark leaderboard captured from https://www.vals.ai/benchmarks/hlab and updated by Vals on October 12, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.
Harvey's Legal Agent Benchmark is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.
Snapshot
The published Harvey's Legal Agent Benchmark snapshot places Muse Spark 1.2 first at 25.42%. The third row is 2.50 points behind. The broader top-10 range is 9.58 points, so many of the published results sit in a relatively narrow band.
79 models have been evaluated on Harvey's Legal Agent Benchmark. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Harvey's Legal Agent Benchmark is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Harvey's Legal Agent Benchmark
Year
2026
Tasks
Legal agent work across documents, spreadsheets, presentations, and files
Format
Accuracy score
Difficulty
Professional legal workflow automation
BenchLM mirrors the public Vals AI Harvey's Legal Agent Benchmark leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.
Freshness and provenance
Version
Harvey's Legal Agent Benchmark 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Harvey's Legal Agent Benchmark measure?
Tests an agent's ability to complete legal work using documents, spreadsheets, presentations, and file-system tools
Which model leads the published Harvey's Legal Agent Benchmark snapshot?
Muse Spark 1.2 currently leads the published Harvey's Legal Agent Benchmark snapshot with 25.42% harvey's legal agent benchmark score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Harvey's Legal Agent Benchmark?
The October 12, 2026 snapshot contains 79 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.