Prime Intellect Releases Prime Agent Coding Harness
Company presents the harness as token-efficient for autonomous coding work.
Prime Intellect posted about Prime Agent, described as a self-improving harness for coding and long-running autonomous tasks. The post outlines its design with programmatic tool calling, context treated as a variable, multi-agent messaging, and a self-modifiable state. Company account states the harness is general-purpose rather than tuned to one benchmark. Separate posts from the same account report performance on ARC-AGI-3 and note gains versus other harnesses on multiple models. Research engineer at the company highlighted collaboration among named contributors and internal use. Other accounts shared the announcement and commented on the results shown in attached images.
Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.
Reactions
93 postsPI 
Prime Intellect@PrimeIntellectPrime Agent is a general-purpose coding harness On ARC-AGI-3, it scores 95.5%, surpassing the human-expert baseline, but the gain is not benchmark-specific. We see major improvements across models when compared to their proprietary harnesses:
- likes: 2.1K
- replies: 78
- bookmarks: 1K
- reposts: 173
AC CR Cameron R. Wolfe, Ph.D.@cwolferesearch@PrimeIntellect insane
- likes: 6
- replies: 0
- bookmarks: 0
- reposts: 0
FB Florian Brand@xeophonTurns out when you get super smart people (@a1zhang @sethkarten @omouamoua and @kevinjosethomas) to collab, you simply get SOTA. Without even trying for specific evals, it just works. Been using this internally for a while, it’s great! Really pushes models forward
- likes: 482
- replies: 14
- bookmarks: 145
- reposts: 22
FB Florian Brand@xeophonAlso, huge shoutouts to @patience_cave and @myainotez who both ran their evals on a short notice to show the efficiency and capability gains of prime-agent without us tailoring the harness to their evals
- likes: 21
- replies: 3
- bookmarks: 3
- reposts: 0
AZ alex zhang@a1zhangReally excited to announce Prime Agent, an RLM-native coding TUI! I've been really focused on RLM research and properties of similar harnesses, but ultimately it's all about bringing out stronger capabilities to users. This agent is really good despite not having any models…
- likes: 997
- replies: 33
- bookmarks: 422
- reposts: 66
