gpt-6POST

GPT-6 Astra underperforms equal-weight benchmark in Quant-Arena test

In Quant-Arena's 2024 test, GPT-6 Astra returned 10.7%, trailing the 17.1% equal-weight benchmark and even a buy-everything strategy.

Rose Carter

@Bisma7465Archived 5 days agoOriginally posted 2026.10.23

Rose Carter @Bisma7465Originally posted 2026.10.23

Open original on X

Quant-Arena makes one thing clear: LLMs still have a long way to go in quant research. GPT-6 Astra returned 10.7% in 2024, while the equal-weight benchmark delivered 17.1%. Even buying every stock performed better. Einsia’s Kepler-1 scored 77.56 against the best model’s 51.29, showing just how much ground frontier models still need to cover.

This post quotes another post: x.com/einsiaai/status/2113276693108408812

x.com/bisma7465/status/2113453849482760564Archived 2026.10.24 10:09