MIT Releases FinanceGym Dataset; Kimi and Qwen Fail Tests
MIT_CSAIL · x · 2026-08-25
MIT CSAIL researchers released "FinanceGym," the largest dataset for finance agents yet, featuring over 59K samples across 37 roles and 40 apps/sites. The dataset addresses the challenge of locked private corporate data in finance agent training. Benchmarks show that even advanced models like Kimi K3 and Qwen 3.8 Max failed to pass its test set.
More from Research
- Optima now supports benchmarking Qwen3.8 27B — ArtificialAnlys · 2026-08-25
- PhysCaP: Physics-Informed Agent for Robotic Manipulation — Chen-Yu Lin · 2026-08-25
- Who Verifies Agents workshop at NeurIPS 2026: submissions due Aug 29 — abeirami · 2026-08-25
- Dynamic Evaluation author: Test-time adaptation holds untapped potential in frontier models — MannyKayy · 2026-08-25
- Claude accelerates protein design with lab-validated results on 14 of 15 targets — thione · 2026-08-25
- Master's Thesis on Model Simplification in Value-based Deep RL Wins Freudenberg Award — Jan_R_Peters · 2026-08-25