DeepFinance: leaderboard

Julius Research's private-equity and investment-banking benchmark: 21 deal tasks that take an analyst 20 to 30 hours, each with a data room of about 19 files, graded against about 84 expert-written rubric criteria on output, workflow and judgment, so sourced and auditable steps earn credit even when a final number is wrong. Each model runs in its vendor's agent harness (Claude Code, Codex, opencode and others) with web search on; the score is the mean share of criteria met over 10 attempts per task. The page does not name the grader model.

Metric: Mean rubric score (%). Source: www.juliusresearch.ai. Saturation forecast: Around September 2027. 28 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Fable 5.2Max62#1
2GPT-6.2 AstraMax61.4#4
3GPT-6 AstraMax57.6#10
4Gemini 4 CarbonMedium56.6#5
5Claude Opus 5.5Max54.8#9
6Claude Sonnet 5.5Max54.6#24
7GPT-6.1 SolMax54.3#15
8Muse Wonder FullMax54.2#21
9GPT-6.2 SolMax51.9#12
10GPT-6.1 TerraMax47.2#25
11Claude Fable 5.1Max45.6#20
12Muse WonderMax44.2#34
13Gemini 4 ArgonHigh41.4#18
14Claude Haiku 5.5Max40#75
15Muse Spark 1.3Max37#48

No result here: #3 Gemini 4 Neon.

Interactive version: theaggregate.ai/benchmark?slug=deepfinance · How It Works · Data refreshed daily, snapshot 2026-11-20.