MIT Releases Largest Finance Agent Dataset; Top Models Fail Tests
qinzytech · x · 2026-08-25
MIT researchers have released FinanceGym, the largest dataset for finance agents, which is 10x larger than the current biggest. It spans 37 finance roles and 40 apps/websites. Notably, even top-tier models like Kimi K3 and Qwen 3.8 Max failed on the test set.
More from coding & agent
- Build a voice agent with LangGraph and ElevenLabs — dl_weekly · 2026-08-25
- Chroma launches Foundation as unified agent memory infrastructure — altryne · 2026-08-25
- Foundation launches as 'memory infrastructure' for $30/month — altryne · 2026-08-25
- Agents can navigate platforms with poor UX — nijfranck · 2026-08-25
- Don't Just Write Prompts, Optimize Them: MLflow Supports DSPy and Others — Odd-Situation6749 · 2026-08-25
- Factory Droid outperforms Claude Code in same-model agent test — matanSF · 2026-08-25