Quant-Arena benchmarks LLM agents on end-to-end quant research
Original title: Quant-Arena: Benchmarking LLM Agents for Quantitative Trading Research
Nine frontier models were scored; GPT-6 Astra led, but none beat equal-weight A-shares in 2024.
Quant-Arena is a public benchmark and leaderboard for whether LLM agents can run a full quantitative-research loop on real A-share data. A model builds features and stock-scoring code, gets feedback only on 2022--2023, and freezes one strategy for a sealed 2024 test. Scoring emphasizes realized excess return after trading constraints, costs, and simulated market impact, plus signal quality and feature diversity. The leaderboard is open to any model or agent harness.
- Feedback is limited to 2022--2023; 2024 is a sealed assessment year.
- A cash-account evaluator includes costs and order-book market impact.
- GPT-6 Astra scored highest among the nine frontier models tested.
- No tested model beat the equal-weight market in 2024.
From the original article
1 min read at huggingface.co
Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual trading decisions, leaving open whether large language models (LLMs) can complete this loop and produce strategies that work beyond the research period.
By Houde Qian, Zheng Jiang, Wenyi Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.