Prime Agent: A Self-Improving RLM Harness
Prime Agent is pitched as an open-source harness that lets coding agents carry state, spawn subagents, and recover across long runs.
The paper says its persistent IPython REPL, Continual Harness, and daemon-backed sessions are meant to separate harness breakdowns from model limits. It reports ARC-AGI-3 RHAE Best@1 rising from 30% to 95.5%. The authors also claim parity or better results across long-context coding, GPU-kernel generation, emulator construction, nanoGPT speedruns, and Factorio progression. HF Daily Papers' note
The paper says its persistent IPython REPL, Continual Harness, and daemon-backed sessions are meant to separate harness breakdowns from model limits. It reports ARC-AGI-3 RHAE Best@1 rising from 30% to 95.5%. The authors also claim parity or better results across long-context coding, GPU-kernel generation, emulator construction, nanoGPT speedruns, and Factorio progression. HF Daily Papers' note
score 5