Julius Research

Julius Research is an applied AI research lab based in San Francisco, CA. We’re the creators of Julius (AI Data Scientist with 2M+ users) and used by Harvard Business School and companies like Nvidia.

DeepFinanceBench

An ultra long-horizon benchmark for measuring models on real-world Private Equity and Investment Banking workflows that would take a human professional between 20 and 30 hours to complete.

tasks: 21
average human time: 27.1 hours per task
rubric criteria per task: 84
files per data room: 19

Results

Mean score out of 100 across 20 tasks, 10 attempts each. Higher is better.

GPT-6 Astra (xhigh)55.5 ±8.0
Claude Sonnet 5.5 (max)54.5 ±9.5
GPT-6.1 Sol (xhigh)54.1 ±7.8
Claude Opus 5.5 (max)50.2 ±9.2
Claude Fable 5.1 (max)45.7 ±5.4
GPT-6.1 Terra (xhigh)44.3 ±6.9
Gemini 4 Argon (high)37.9 ±5.3
Kimi K3.1 Pro (max)35.5 ±4.2
Grok 4.7 (xhigh)34.4 ±4.5
Qwen 3.8 Max (xhigh)30.8 ±4.0
DeepSeek V4.1 Flash (max)27.9 ±3.9
GLM 5.3 (max)26.3 ±4.5
GPT-6 Luna (xhigh)25.1 ±4.8
Mistral Large 4 (high)21.9 ±4.6
DeepSeek V4 Pro (max)20.7 ±4.2

010203040506070

Score (0–100)

Results: mean score out of 100 with its 95% confidence interval, best first
Rank	Model	Lab	Score (0–100)	95% confidence interval
1	GPT-6 Astra (xhigh)	OpenAI	55.5	±8.0
2	Claude Sonnet 5.5 (max)	Anthropic	54.5	±9.5
3	GPT-6.1 Sol (xhigh)	OpenAI	54.1	±7.8
4	Claude Opus 5.5 (max)	Anthropic	50.2	±9.2
5	Claude Fable 5.1 (max)	Anthropic	45.7	±5.4
6	GPT-6.1 Terra (xhigh)	OpenAI	44.3	±6.9
7	Gemini 4 Argon (high)	Google	37.9	±5.3
8	Kimi K3.1 Pro (max)	Moonshot AI	35.5	±4.2
9	Grok 4.7 (xhigh)	SpaceXAI	34.4	±4.5
10	Qwen 3.8 Max (xhigh)	Alibaba	30.8	±4.0
11	DeepSeek V4.1 Flash (max)	DeepSeek	27.9	±3.9
12	GLM 5.3 (max)	Z.ai	26.3	±4.5
13	GPT-6 Luna (xhigh)	OpenAI	25.1	±4.8
14	Mistral Large 4 (high)	Mistral	21.9	±4.6
15	DeepSeek V4 Pro (max)	DeepSeek	20.7	±4.2

Error bars show the 95% confidence interval across tasks. Updated Nov 4, 2026.

Read the post: DeepFinanceBench: Can AI diligence a $100MM deal?