Olas releases forecasting model and overnight autonomous Pearl agent
Olas open-sourced Olas-Predict-R1-14B, a forecasting model with about 20% lower error than its base model, and gave the Pearl Success Agent overnight autonomy to reply and fix bugs.
This week in Pearl — what's new: → Olas-Predict-R1-14B, an open-weight forecasting model fine-tuned on Olas agent data, cuts forecast error by ~20% vs. its base model and performs on par with GPT-4.1 in a six-week evaluation https://olas.network/blog/meet-olas-predict-r1-14b → The Pearl Success Agent loop now runs overnight: it's designed to reply to you without waiting for a person's review, and to write and review bug fixes so engineers only sign off on the final result https://olas.network/blog/pearl-success-overnight → Pearl Success Agent helped unblock 5 users this week Set up your first agent in minutes: http://pearl.you