📰1400Read later0Read MoreNanoGPT Speedrun FrontierHacker News·2 months ago#79J1dxBY#primeintellect#nanogpt#speedrun#optimizer#photo+1 more🧰Tag tools✨Add tagCompare 153 autonomous runs from 18 frontier models on the nanoGPT optimizer speedrun.15s
🖼️2530Read later0Read MorePrime Agent: A self-improving RLM agentHacker News·3 months ago#cpa3Yh5r#primeintellect#agent#prime#harness#agents+4 more🧰Tag tools✨Add tagPrime Agent is our open-source, self-improving coding harness built around two abstractions: the Recursive Language Model (RLM) and the Continual Harness. With Opus 5, it achieves 95.5% on ARC-AGI-3, surpassing the reported human expert baseline.15s
