PRBench Legal

Professional reasoning in legal practice, graded by practitioners. Scale AI.

Published by

Scale AI SEAL

Category

Knowledge

Index weight

0.5

Models

33

Data as of

9 Sept 2026

Scale AI SEAL Leaderboards.

1. 1Muse Spark 1.157.0%
2. 2Claude Fable 552.6%
3. 3Muse Spark52.3%
4. 4Claude Opus 4.652.3%
5. 5Claude Fable 5.151.6%
6. 6GPT-5.6 Sol50.5%
7. 7GPT-5 Pro49.9%
8. 8o3-pro49.7%
9. 9GPT-5.149.3%
10. 10GPT-549.0%
11. 11Gemini 3.8 Flash48.9%
12. 12o348.6%
13. 13GPT-6 Astra48.4%
14. 14GPT-5.2 Pro45.4%
15. 15GPT-5.444.4%

SearchAll providersAlibabaAnthropicDeepSeekGoogleMetaMistral AIMoonshot AIOpenAIOpen weights33 of 33

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Muse Spark 1.1Meta	57.0%	Muse Spark 1.1	64.9	–
2	Claude Fable 5Anthropic	52.6%	claude-fable 5	70.4	–
3	Muse SparkMeta	52.3%	–	65.4	–
4	Claude Opus 4.6Anthropic	52.3%	–	55.3	–
5	Claude Fable 5.1Anthropic	51.6%	Fable 5.1	70.0	–
6	GPT-5.6 SolOpenAI	50.5%	–	65.5	–
7	GPT-5 ProOpenAI	49.9%	–	60.7	–
8	o3-proOpenAI	49.7%	–	54.1	–
9	GPT-5.1OpenAI	49.3%	–	55.2	–
10	GPT-5OpenAI	49.0%	gpt-5	59.5	–
11	Gemini 3.8 FlashGoogle	48.9%	Gemini 3.8 Flash	62.4	–
12	o3OpenAI	48.6%	o3	60.3	–
13	GPT-6 AstraOpenAI	48.4%	GPT 6 Astra	69.9	–
14	GPT-5.2 ProOpenAI	45.4%	–	–	–
15	GPT-5.4OpenAI	44.4%	–	58.8	–
16	Claude Opus 4.5Anthropic	44.2%	–	60.6	–
17	Gemini 3.1 ProGoogle	44.0%	–	63.9	–
18	Kimi K2.5Moonshot AIopen	43.8%	kimi-k2.5	56.1	–
19	Gemini 2.5 ProGoogle	41.4%	–	54.1	–
20	Gemini 2.5 FlashGoogle	41.0%	–	48.8	–
21	Kimi K2 ThinkingMoonshot AIopen	40.9%	–	52.8	–
22	Claude Sonnet 4.5Anthropic	40.8%	–	52.3	–
23	Gemini 3 ProGoogle	40.6%	–	61.6	–
24	gpt-oss-120bOpenAIopen	40.2%	–	47.6	–
25	Mistral Medium (latest)Mistral AI	39.5%	–	–	–
26	Qwen Qwen3 235B A22bAlibabaopen	38.3%	2507	–	–
27	o4-miniOpenAI	38.1%	–	54.9	–
28	DeepSeek R1DeepSeekopen	36.6%	0528	48.9	–
29	GPT-4.1OpenAI	36.5%	gpt-4.1	47.3	–
30	Kimi K2Moonshot AIopen	36.4%	–	48.8	–
31	Claude Opus 4.1Anthropic	34.0%	–	52.6	–
32	GPT-4.1 miniOpenAI	30.4%	–	44.5	–
33	Llama 4 Maverick BasicMeta	24.8%	–	38.8	–