MACR: confidence-guided relay multi-agent framework cuts token overhead 6x
Xu Liu · hf · 2026-11-03
A paper from Zhejiang University, NTU and HKU introduces MACR (Multi-Agent Confidence-guided Relay), replacing group-thinking multi-agent collaboration with a relay topology where reasoning ownership dynamically transfers and subsequent agents inherit and refine an evolving trajectory. It decides when to relay and which agent should continue via confidence-guided triggering plus a bidding mechanism. Across four benchmarks it beats the strongest single-model strategies (+3.3% on MathVista, +3.9% on MMMStar) while matching multi-strategy methods with 6.0x less token overhead on HallusionBench and MMMStar.
More from coding & agent
- User says Codex wiped systemd and containers instead of rebooting, costing $10,000+ — TheZachMueller · 2026-11-03
- Coder pitches an 'AI Operating Layer' to govern agents without losing control — CodeByPoonam · 2026-11-03
- DHH: programmers may be worse at AI coding than non-coders — '100%' — 2C_ornot2C · 2026-11-03
- Running a family trip with two agents: Teamily AI in action — thetripathi58 · 2026-11-03
- Dev says AI coding is collapsing to intent plus privileged SOR access — curious_vii · 2026-11-03
- Developer Open-Sources Cosmion: One-Command Image Generation for Claude Code and Other Agents — TheAverageJedi · 2026-11-03