Google’s Gemini 4 Neon tripled its coding score without changing the model
Google’s new Gemini 4 Neon resolved 12.8% of the SWE-sweep coding benchmark using the mini-SWE harness, but that score jumped The post Google’s Gemini 4 Neon tripled its coding score without changing the model appeared first on The New Stack .
Google's Gemini 4 Neon model has shown a significant increase in coding performance, tripling its score on the SWE-sweep benchmark when run within the Antigravity harness, jumping from 12.8% to 39.0%. This substantial improvement has raised questions about the extent to which the underlying model contributes to coding-agent performance and how much is influenced by the harness managing its tools, context, and execution.
The company's testing did not isolate these factors, leaving uncertainty about which changes contributed most to the SWE-sweep improvement.
Brief written by urgent.news from The New Stack's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- I gave my local LLM live web search for free, and it replaced the one thing I still used ChatGPT and Gemini for xda-developers.com
- Today I learned a single Google Gemini query uses about five drops of water, but that's not the whole truth techradar.com
- Google upgrades Gemini’s agentic Task mode with new model, e-commerce features siliconangle.com
- Google is giving millions of Gemini users a free personal AI agent androidpolice.com
- Google rebrands its Gemini Spark AI agent as Tasks and makes it free in six countries; in the US it can call businesses and buy items with Google Pay (Emma Roth/The Verge) theverge.com
- Gemini 4 Neon is ready to get some work done for AI Pro power users androidauthority.com
- Google rolls out Gemini 4 Neon, with "fast end-to-end task completion" across coding, knowledge work, and science, in the API, Antigravity, and the Gemini app (Koray Kavukcuoglu/Google) blog.google
- Gemini 4 Neon costs $3/1M input tokens and $15/1M output tokens; the model sits between Hydrogen and Carbon and replaces Argon in the Gemini app (Abner Li/9to5Google) 9to5google.com