Atlas

Benchmarks

← All benchmarks

Harvey's Legal Agent Benchmark - Arbitration International Dispute Resolution

Professional Work · 2026-05-06

The Arbitration International Dispute Resolution split of Vals' archived v1 Harvey Legal Agent Benchmark held-out evaluation. Its task score averages two judges' all-criteria-pass decisions rather than counting binary task successes.

Top models (higher is better)

ModelScore
GLM-5.320.0
MiMo-V2.6-Flash20.0
Muse Spark 1.220.0
Muse Spark 1.320.0
Muse Spark Open20.0
Muse Wonder Full20.0
Haiku 4.50.0
Claude Haiku 5.50.0
Opus 4.70.0
Opus 4.80.0
Claude Opus 5.50.0
Claude Opus 50.0
Sonnet 4.60.0
Claude Sonnet 5.50.0
Sonnet 50.0
DeepSeek V4.1 Flash 10170.0
DeepSeek V4.1 Flash0.0
DeepSeek-V4-Flash-07310.0
DeepSeek V4 Pro 08130.0
DeepSeek-V4-Pro0.0
Ember-10.0
Gemini 3.1 Flash-Lite Preview0.0
Gemini 3.1 Pro Preview0.0
Gemini 3.5 Flash0.0
Gemini 3.5 Flash-Lite0.0
Gemini 3.6 Flash0.0
Gemini 3.7 Flash0.0
Gemini 3.8 Flash0.0
Gemini 3 Flash Preview0.0
Gemini 4 Argon0.0
GLM-5.10.0
GLM-5.20.0
GLM-5.3 Flash0.0
GPT-5.4 Mini0.0
GPT-5.4 Nano0.0
GPT-5.40.0
GPT-5.50.0
GPT-5.6 Luna0.0
GPT-5.6 Sol0.0
GPT-5.6 Terra0.0
Loading Atlas data…