Legora BAR

Hundreds of private cases drafted by Legora’s law-firm partners and legal engineers, across 30 practice areas and seven task families, from a short NDA redline to a full statement of claim or transfer-pricing master file. Each model works every case three times inside Legora’s own agent harness, in a file room holding the matter’s documents, playbooks, and templates, and an LLM judge scores the work against importance-weighted rubrics written by legal professionals. Legora publishes each result relative to the average of the 17 models it ran.

Last updated · Source leaderboard checked daily

Main Insights

Results

Legora BAR results, ranked by score (quality vs average), highest first. Last updated .
RankModelProviderScore (quality vs average)Cost (USD / task)Duration (seconds / task)
1Claude Opus 5.5Anthropic106.0%$251.00123
2Muse Wonder 1.0Meta105.0%$50.0032
3Claude Fable 5.1Anthropic104.0%$263.0079
4Gemini 3.8 FlashGoogle103.0%$20.0083
5Mistral Large 4Mistral103.0%$132.00330
6Claude Sonnet 5.5Anthropic102.0%$150.00164
7Kimi K3Moonshot AI102.0%$51.0082
8GPT-6.1 SolOpenAI102.0%$44.0078
9Grok 4.7xAI102.0%$79.0092
10GLM 5.3Z.ai102.0%$52.0083
11Muse Spark 1.3Meta101.0%$18.0034
12DeepSeek V4.1 FlashDeepSeek100.0%$5.0048
13Gemini 4 ArgonGoogle99.0%$130.00121
14GPT-6 AstraOpenAI96.0%$432.00100
15GLM 5.3 FlashZ.ai94.0%$7.0093
16DeepSeek V4 ProDeepSeek93.0%$9.0056
17GPT-6 LunaOpenAI88.0%$7.00101