Spine-Branch Framework Boosts Multi-Agent Success by 16.5%
ZhiyuChen4 · x · 2026-08-25
Scale AI researchers introduced Spine-Branch Coordination, a framework for multi-agent computer use that addresses the physical impossibility of merging live VM states.
- Spine: Maintains a single, persistent VM for the main task flow and costly state (e.g., active sessions).
- Branch: Executes parallel sub-tasks to gather artifacts, then discards the VMs, transferring only mergeable data.
By avoiding VM merging and reconstruction, this approach enables efficient scaling. Experiments on 200 long-horizon tasks across three CUA backbones show that Spine-Branch improves success rates by 6.0–16.5% while reducing per-task costs by 34–70%.
More from coding & agent
- Claude Code 2.1.245 internals: a swarm of internal model names added and removed — ClaudeCodeLog · 2026-08-25
- Claude Code 2.1.245 patch fixes startup crash on glibc 2.44 distros — ClaudeCodeLog · 2026-08-25
- Claude Code 2.1.245 Fixes Crash on Linux glibc 2.44 — ClaudeCodeLog · 2026-08-25
- gitea-mcp: MCP Server with 186 Tools for Gitea API Integration — modelcontextprotocol · 2026-08-25
- Long-Horizon Agent Dev Pain: Not Enough Time to Run Full Rollout — agihouse_org · 2026-08-25
- Terminal-Bench 3.0: Top Model Score Plummets to 43.5% as Agents Fail to Fix Root Causes — ajratner · 2026-08-25