Antigravity CLI tops AA coding index
Google's Antigravity CLI paired with Gemini 4 Neon has claimed the #1 spot on the Artificial Analysis Coding Agent Index while posting massive gains on Meta's SWE-sweep benchmark. The leap underscores how dedicated agent harnesses and subagent orchestration multiply raw foundation model coding capabilities.
Raw model intelligence hits diminishing returns without specialized scaffolding; the harness is now where coding agent wars are won. Google DeepMind's architectural focus on AGY CLI proves that intelligent tooling and subagent fan-out can push mid-tier models past frontier giants.
- –Neon via AGY CLI captured the top rank on the Artificial Analysis Coding Agent Index, outperforming standalone frontier configurations.
- –On Meta's SWE-sweep benchmark—which requires unprompted bug discovery in large repos—the harness scored 39.0%, dwarfing mini-SWE baselines.
- –Antigravity's subagent coordination allows deep codebase exploration without blowing out single-session context windows.
- –Mid-tier models like Neon matching or beating frontier flagships signals a major cost-efficiency shift for autonomous software engineering.
DISCOVERED
25m ago
2026-11-22
PUBLISHED
1h ago
2026-11-21
RELEVANCE
AUTHOR
_mohansolo