FrontierFinance

Two hundred twenty open-ended investment research questions written by finance experts, from screening and company research to data extraction, catalysts, and earnings. The agent runs in Samaya’s version of Vals AI’s open-source finance agent harness, with SEC EDGAR search, market prices, web search, page reading, and a calculator. Three LLM judges check each answer against an average of 52 expert-written, source-attributed rubric items, and the score is the share of items satisfied.

Last updated · Source leaderboard checked daily

Main Insights

Results

FrontierFinance results, ranked by score (rubric items satisfied), highest first. Last updated .
RankModelProviderScore (rubric items satisfied)Cost (USD / task)Duration (seconds / task)
1Claude Fable 5.1Anthropic55.9%$6.86268
2Muse Spark OpenMeta53.7%$1.38234
3Muse Wonder 1.0Meta52.6%$3.35292
4Kimi K3.1 ProMoonshot AI51.7%$1.56449
5Claude Opus 5.5Anthropic50.9%$1.87148
6Gemini 3.8 FlashGoogle50.0%$1.21247
7Claude Fable 5Anthropic49.2%$4.06165
8GPT-5.6 SolOpenAI46.8%$3.03171
9Ling 3.0 Flash FininclusionAI46.6%$0.050318
10Kimi K3Moonshot AI46.4%$0.90336
11Gemini 3.6 FlashGoogle46.3%$2.41164
12Claude Opus 4.8Anthropic45.0%$2.61156
13GPT-5.5OpenAI43.5%$2.80233
14GLM 5.2Z.ai42.8%$0.63297
15DeepSeek V4 ProDeepSeek40.5%$0.68203
16Kimi K2.6Moonshot AI32.3%$0.95201
17Gemini 3.1 ProGoogle30.5%$1.72222