PRBench Finance

Professional reasoning in finance, 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 1.155.0%
2. 2Claude Fable 553.9%
3. 3Claude Opus 4.653.3%
4. 4Muse Spark52.4%
5. 5GPT-551.3%
6. 6GPT-5 Pro51.1%
7. 7Claude Fable 5.150.8%
8. 8GPT-5.6 Sol50.5%
9. 9Gemini 3.8 Flash49.5%
10. 10o3-pro49.1%
11. 11GPT-5.148.0%
12. 12o347.7%
13. 13GPT-6 Astra47.5%
14. 14Kimi K2.546.5%
15. 15GPT-5.2 Pro46.3%

SearchAll providersAlibabaAnthropicDeepSeekGoogleMetaMistral AIMoonshot AIOpenAIOpen weights33 of 33

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Muse Spark 1.1Meta	55.0%	Muse Spark 1.1	64.9	–
2	Claude Fable 5Anthropic	53.9%	claude-fable 5	70.4	–
3	Claude Opus 4.6Anthropic	53.3%	–	55.3	–
4	Muse SparkMeta	52.4%	–	65.4	–
5	GPT-5OpenAI	51.3%	gpt-5	59.5	–
6	GPT-5 ProOpenAI	51.1%	–	60.7	–
7	Claude Fable 5.1Anthropic	50.8%	Fable 5.1	70.0	–
8	GPT-5.6 SolOpenAI	50.5%	–	65.5	–
9	Gemini 3.8 FlashGoogle	49.5%	Gemini 3.8 Flash	62.4	–
10	o3-proOpenAI	49.1%	–	54.1	–
11	GPT-5.1OpenAI	48.0%	–	55.2	–
12	o3OpenAI	47.7%	o3	60.3	–
13	GPT-6 AstraOpenAI	47.5%	GPT 6 Astra	69.9	–
14	Kimi K2.5Moonshot AIopen	46.5%	kimi-k2.5	56.1	–
15	GPT-5.2 ProOpenAI	46.3%	–	–	–
16	Claude Opus 4.5Anthropic	46.2%	–	60.6	–
17	GPT-5.4OpenAI	45.6%	–	58.8	–
18	gpt-oss-120bOpenAIopen	43.8%	–	47.6	–
19	Claude Sonnet 4.5Anthropic	43.8%	–	52.3	–
20	Kimi K2 ThinkingMoonshot AIopen	43.4%	–	52.8	–
21	Gemini 3.1 ProGoogle	41.9%	–	63.9	–
22	Mistral Medium (latest)Mistral AI	39.4%	–	–	–
23	o4-miniOpenAI	39.2%	–	54.9	–
24	Gemini 3 ProGoogle	39.2%	–	61.6	–
25	Qwen Qwen3 235B A22bAlibabaopen	39.1%	2507	–	–
26	Gemini 2.5 ProGoogle	38.9%	–	54.1	–
27	Gemini 2.5 FlashGoogle	38.4%	–	48.8	–
28	Kimi K2Moonshot AIopen	38.3%	–	48.8	–
29	Claude Opus 4.1Anthropic	35.1%	–	52.6	–
30	GPT-4.1OpenAI	34.3%	gpt-4.1	47.3	–
31	DeepSeek R1DeepSeekopen	32.7%	0528	48.9	–
32	GPT-4.1 miniOpenAI	30.4%	–	44.5	–
33	Llama 4 Maverick BasicMeta	22.4%	–	38.8	–