A-Evolve: Universal Infrastructure for Self-Improving Agents

A-Evolve is an open-source framework that positions itself as “the PyTorch for Agentic AI” — an infrastructure that evolves any agent across any domain using any evolution algorithm with zero human intervention. Published as an academic position paper (arXiv 2602.00359) titled “Agentic Evolution is the Path to Evolving LLMs.”

Benchmark results with zero manual harness engineering:

• MCP-Atlas: 79.4% (🥇 #1, +3.4pp)
• SWE-bench Verified: 76.8% (~#5, +2.6pp)
• Terminal-Bench 2.0: 76.5% (~#7, +13.0pp)
• SkillsBench: 34.9% (#2, +15.2pp)
• ARC-AGI: 12.3% (🥇 #2 Community Leaderboard, +2.2pp)
• OSWorld: 69.6%

Usage is just 3 lines of Python — provide a Base Agent, call ae.Evolver(agent="./my_agent", benchmark="swe-verified"), get back a SOTA agent. Supports any evolution algorithm and any domain.