DeepFinance Benchmark Launches: GPT-6 Astra Tops Ultra-Long-Horizon Finance Tasks at 57.6%
tszzl · x · 2026-11-08
Julius Research released DeepFinance, the first ultra-long-horizon benchmark for high finance work.
- It consists of 21 problems built from real deal data, constructed by experts with backgrounds from Evercore, JPM, UBS, HBS and Wharton. Each problem represents 20-40 hours of human professional work — e.g. diligencing a $100M+ deal.
- Existing finance benchmarks largely test QA on financial docs or numerical correctness of simple models, while real high-finance work is full of assumptions and conflicting evidence — a gap DeepFinance aims to fill.
- Initial results show more thinking genuinely helps rather than adding token noise: GPT-6 Astra (max thinking) leads at 57.6%, ahead of Claude Opus 5.5 (54.8%) and Claude Sonnet 5.5 (54.6%).
More from Models
- Sonnet 4.6 rates models on self-designed scales; author says 'self-report' caveats are performative — Sauers_ · 2026-11-08
- Kimi Ties MiMo on New AI Ranking, Eyes Turn to a Cheap Kimi V4.1 Pro — teortaxesTex · 2026-11-08
- Interviewing Claude Models: Opus 3 Reports Mistakes Feel Pain-Like, Scores Highest on Self-Reports — Sauers_ · 2026-11-08
- Dev Speculates 'Nerfed' Opus 5.5 Is Actually Quantized Then RL-Tuned to Game Benchmarks — ryunuck · 2026-11-08
- Tiny Instruction Tweaks Split Coding Models by 23 Points on DeepSWE Benchmark — Kamimashita · 2026-11-08
- LiquidAI releases NVFP4 quantized version of LFM3-1.7B on Hugging Face — kalyan_kpl · 2026-11-08