Ant Group's Mara Chain turns rejected candidates into stepping stones for AI auto-evolution
antgroup · hf · 2026-10-10
Ant Group introduced Mara Chain, a refinement procedure for auto-optimizing prompts, skills, harnesses, and code. Instead of discarding rejected candidates in propose-evaluate-select loops, it iteratively refines them with accumulated evidence, with fixed chain depth and Pareto-filtered Top-N selection.
- AppWorld skill optimization: up to 20.5% relative gains over GEPA/ACE/SkillOpt-Lite, reaching target scores with 65.5% fewer rollouts than GEPA
- TerminalBench 2.1 harness optimization: +20.2 and +22.5 points pass rate over AHE and Meta-Harness
- MuSiQue retrieval pipeline: +0.104 nDCG@10 and +0.131 Recall@10 over a hand-written baseline
More from coding & agent
- A daily prompt that makes AI refactor your logs and errors for better debugging — DanielLockyer · 2026-10-10
- Microsoft pairs Surface Laptop Ultra with agent sandbox MXC and AI routing in Windows — pstAsiatech · 2026-10-10
- Open-source creative harness Voyager adds 8 languages, drives Blender, Resolve and 100+ tools — ravisparikh · 2026-10-10
- Claude Managed Agents finds 21 real bugs in 18k lines of code in 5 minutes for $8.36 — daniel_mac8 · 2026-10-10
- Testing an AI agent to book a hotel failed — and it still required manually entering credit card info — NielsRogge · 2026-10-10
- AI now writes 52% of merged code, but DX researcher says stop treating AICP as a KPI — rseroter · 2026-10-10