Antigravity harness pushes Gemini to #1 on coding agent index with huge SWE-sweep jump
rseroter · x · 2026-11-22
Google's Antigravity team says the harness is now the best setup for Gemini on complex tasks: Neon using the AGY CLI tops the AA Coding Agent Index and posts a massive jump over mini-swe on SWE-sweep, a bug-finding/fixing eval on large codebases.
The shared benchmark table (Gemini 4 Neon vs Claude Opus 5.5 vs GPT-6.2 Sol) shows Neon leading on AA Index (61), Vals Index (70.9%), and CursorBench (65.4%). On SWE-sweep the gap is stark: 39.0% with the provider harness vs 10.1% and 8.3% for rivals, and Neon still leads with the Mini-SWE harness (12.8%). The team says the harness improvement is larger than native harnesses of other frontier models, with more updates coming next week.
More from coding & agent
- Open-source AI video editor Scarlet adds Gemini support, frame annotation and lip-sync fixes — MatthewBerman · 2026-11-22
- Harrison Chase: buying Copilot won't transform teams without a knowledge factory — hwchase17 · 2026-11-22
- OpenAI's $200 Codex Edition ModRetro handheld turns prompts into playable cartridge games — shaunralston · 2026-11-22
- Aiden Bai doubles down: "** agents!!!" — aidenybai · 2026-11-22
- Cognition CEO Aiden Bai says he never reads the code, just watches agent test videos — aidenybai · 2026-11-22
- Long-running agents fail by re-litigating old decisions, not by crashing — ClickOk5811 · 2026-11-22