Alpheva Finance: Financial Reasoning Benchmark

Financial reasoning under incomplete information.

Can a model make defensible financial decisions when information is incomplete?

GPT-5.5(OpenAI)

79.4%

Claude Opus 4.7(Anthropic)

78.9%

DeepSeek V4 Pro(DeepSeek)

77.5%

Gemini 3.1 Pro(Google)

76.8%

Kimi K2.6(Moonshot AI)

75.1%

Gemini 3.8 Flash(Google)

73.9%

Grok 4.3(xAI)

73.6%

GLM-5.1(Zhipu AI)

72.4%

Claude Fable 5(Anthropic)

71.5%

MiniMax M2.7(MiniMax)

69.6%

Claude Opus 5(Anthropic)

68.2%

GPT-5.6 Sol(OpenAI)

67.4%

Grok 4.6(xAI)

66.8%

GPT-5.6 Luna(OpenAI)

65.3%

Muse Spark 1.2(Meta)

64.8%

Claude Sonnet 5(Anthropic)

63.9%

Mean

Overview

Alpheva Finance evaluates how models interpret primary filings, construct defensible assumptions, revise analysis as evidence changes, and communicate uncertainty while following professional accounting standards.

Evaluation focus

The benchmark is organized around the capabilities that determine whether a model can complete the full workflow reliably.

01Primary-source financial analysis
02Logical and defensible assumptions
03Accounting-standard adherence
04Exact calculation accuracy

Protocol

Protocol: Alpheva v0.1
Materials: Filings + workpapers
Scope: Evidence-aware judgment

Model results

Model	Provider	Mean
GPT-5.5	OpenAI	79.4%
Claude Opus 4.7	Anthropic	78.9%
DeepSeek V4 Pro	DeepSeek	77.5%
Gemini 3.1 Pro	Google	76.8%
Kimi K2.6	Moonshot AI	75.1%
Gemini 3.8 Flash	Google	73.9%
Grok 4.3	xAI	73.6%
GLM-5.1	Zhipu AI	72.4%
Claude Fable 5	Anthropic	71.5%
MiniMax M2.7	MiniMax	69.6%
Claude Opus 5	Anthropic	68.2%
GPT-5.6 Sol	OpenAI	67.4%
Grok 4.6	xAI	66.8%
GPT-5.6 Luna	OpenAI	65.3%
Muse Spark 1.2	Meta	64.8%
Claude Sonnet 5	Anthropic	63.9%

Results and methodology

Scores are reported against the protocol and versioned environment defined above. Compare results within this benchmark and metric only.