AFAC2026 Finals: 20K-Contestant Financial AI Contest Opens Million-Sample Dataset
量子位 · wechat · 2026-09-04
AFAC2026, co-founded by 33 institutions including Ant Group, Peking University and CCF, held its finals in Shanghai with 5,027 teams competing for a 1% spot. Four challenges mirror real finance AI pain points: trading-behavior recognition under dynamic markets, complex financial document restoration, automated experiments with sparse feedback, and dynamic memory compression for long-context agents — with judges stressing that token economics and context control, not prompt polish, separate winners. The contest also open-sourced the AFAC million-sample financial AI dataset (14 tasks, 130K eval samples, 2,000+ documents, 130+ winning solutions) from 2023–2025, alongside a million-yuan prize pool, compute support and VC matchmaking.
More from Companies & People
- AI evals engineers' real pay: median band $216k–$337k across 220 postings, Redwood tops at $850k — aryaman2020 · 2026-09-04
- Google researcher warns C++-to-Python shift is creating devs who don't understand computers — Franc0Fernand0 · 2026-09-04
- Max Planck Schools open first Biomedical AI PhD cohort, fully funded for 5 years — kmborgwardt · 2026-09-04
- Nadella says early customers already use Astra on Azure as Altman responds — i_dg23 · 2026-09-04
- Only 33% of Americans Trust AI Financial Advice vs 71% for Human Advisors — mikeflache · 2026-09-04
- MoonBit team raises funding to form a company, hiring for agents and wasm — dotey · 2026-09-04