BlueFin - Synthesis: leaderboard
The 9 held-out synthesis tasks of BlueFin (Kundurthy et al., 2026): the agent builds a financial-model workbook from a natural-language brief through a spreadsheet tool harness. Scored in percent (higher is better) as the weighted share of expert rubric criteria met, checked by an agentic GPT-5.4 judge.
Metric: Criteria-weighted rubric score (%; points met over maximum points of expert rubrics checked by an agentic GPT-5.4 judge for synthesis and manipulation workbooks, answer correctness for interrogation, non-completions count as zero; 20-tool spreadsheet harness, high reasoning effort; 9 held-out tasks building a financial model from a brief). Source: arxiv.org. Saturation forecast: Around May 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (High) | 66.7 | #75 (Claude Opus 4.7) |
| 2 | GPT-5.5 (High) | 59.9 | #48 (GPT-5.5) |
| 3 | Claude Sonnet 4.6 (High) | 53.2 | #158 (Claude Sonnet 4.6) |
| 4 | Gemini 3.1 Pro (Preview) (High) | 50.3 | #91 (Gemini 3.1 Pro (Preview)) |
| 5 | Grok 4.20 (High) | 46.9 | #296 (Grok 4.20) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=bluefin-synthesis · How It Works · Data refreshed daily, snapshot 2026-11-12.