Harvey's Legal Agent Benchmark (Vals AI) leaderboard
- Gemini 4 Argon leads
- 27.1%
- models measured
- 22
- lead over #2
- +2.1 pts
- median
- 11.5%
Leader on Harvey's Legal Agent Benchmark (Vals AI) · read October 22
On October 22, 2026, 22 of our catalog's models had a Harvey's Legal Agent Benchmark (Vals AI) result, and Google Gemini 4 Argon led with 27.1%, ahead of Meta Muse Spark Open at 25.0%.
Legal work with documents, spreadsheets and presentations, done by an agent. The task pass rate, as Vals AI publishes it.
Role in the index
Harvey's Legal Agent Benchmark (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle | 27.1% | $10.55 | 29m 15s | 81.0 | $10.00 | |
| 2 | Muse Spark OpenMeta | 25.0% | $13.63 | 3h 59m | 47.4 | $4.25 | |
| 3 | Muse WonderMeta | 23.8% | $7.08 | 15m 06s | 71.9 | $20.00 | |
| 4 | Grok 4.7xAI | 18.3% | $10.75 | 35m 48s | 66.7 | $6.00 | |
| 5 | Mistral Large 4Mistral AI | 17.5% | $4.18 | 58m 04s | 56.4 | $4.18 | |
| 6 | GLM 5.3Z.ai | 14.6% | $3.99 | 31m 51s | 70.6 | $4.40 | |
| 7 | DeepSeek V4.1 Flash 1017DeepSeek | 14.2% | $0.61 | 22m 08s | 70.3 | $0.60 | |
| 8 | MiMo-V2.6-ProXiaomi | 13.8% | $0.22 | 31m 02s | 65.4 | $0.87 | |
| 9 | MiMo-V2.6-FlashXiaomi | 13.3% | $0.09 | 22m 23s | 54.7 | $0.28 | |
| 10 | Kimi K3Moonshot AI | 12.9% | $4.14 | 42m 36s | 62.3 | $15.00 | |
| 11 | Gemini 3.8 FlashGoogle | 12.1% | $3.66 | 19m 22s | 61.6 | $3.75 | |
| 12 | DeepSeek V4 Pro 0813DeepSeek | 10.8% | $0.80 | 10m 55s | 56.1 | $1.98 | |
| 13 | Claude Sonnet 5.5Anthropic | 10.4% | $16.61 | 1h 13m | 87.3 | $10.00 | |
| 14 | GLM 5.3 FlashZ.ai | 10.4% | $0.32 | 12m 25s | 57.7 | $0.50 | |
| 15 | Claude Opus 5.5Anthropic | 9.2% | $21.38 | 58m 27s | 88.8 | $20.00 | |
| 16 | Hy4 previewTencent | 8.8% | $0.98 | 33m 09s | 44.3 | $2.501 | |
| 17 | GPT-6 AstraOpenAI | 7.1% | $25.46 | 32m 33s | 85.1 | $50.00 | |
| 18 | GPT-6.1 SolOpenAI | 6.3% | $3.54 | 40m 26s | 82.4 | $10.00 | |
| 19 | Claude Haiku 5.5Anthropic | 4.2% | $4.26 | 43m 41s | 70.7 | $0.50 | |
| 20 | GPT-6 LunaOpenAI | 4.2% | $0.30 | 21m 45s | 57.3 | $0.50 | |
| 21 | T InklingThinking Machines | 3.8% | $1.25 | 8m 50s | 31.5 | $4.05 | |
| 22 | T Inkling SmallThinking Machines | 2.9% | $0.18 | 7m 35s | 27.3 | $1.20 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars. Every other column is a figure the board publishes for the same run.