Hebbia Financial Services Benchmark
Professional Work
Hebbia's Financial Services Benchmark is 151 of the hardest private tasks from Hebbia's production finance work: question answering over credit, deal, diligence and equity-research document sets, building or filling Excel models such as comps, LBOs and DCFs, and turning analyst requests into client-ready PowerPoint decks. Agents run in Hebbia's production harness with document retrieval and licensed financial data. A rollout passes only if it covers every expert-calibrated rubric criterion and uses the expected tools and sources; the strict pass rate averages each task's share of three passing rollouts.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 74.6 |
| GPT-6.1 Sol | 73.5 |
| GPT-6 Astra | 72.3 |
| DeepSeek V4.1 Flash | 63.1 |
| Gemini 3.8 Flash | 60.9 |
| Kimi K3 | 60.3 |
| Muse Spark Open | 49.0 |
| GLM-5.3 Flash | 43.7 |