BREAKING
Prime Agent Hits 95.5% on ARC-AGI-3
0
%
Prime Agent Best@1
0
%
Human baseline
0
%
Frontier models solo
Two Core Abstractions
Recursive Language Model
●
Treats context as a variable
●
Delegates via function calls in a REPL
Continual Harness
●
Edits its own prompts and skills
●
Enables run-time self-correction
0
%
Best@3
0
Levels completed
0
Runs
Reaction Is Mixed
Open Source and Available Now
AI NEWS BLITZ
Prime Intellect just released Prime Agent, an open-source self-improving agent harness.