Google released Gemini 4 Neon on 18 November 2026, a mid-tier model in its Gemini 4 family that ranks among the top performers in independent evaluations. Google said the model brings frontier level intelligence with fast task completion across software engineering, knowledge work and science. The model sits between Gemini 4 Hydrogen, built for fast everyday work, and Gemini 4 Carbon, built for the hardest long horizon projects. Google said the model is adapted from Gemini 4 Carbon.
On the Artificial Analysis Intelligence Index, Gemini 4 Neon scored 61 at its highest thinking level, which placed it third overall. Anthropic Claude Fable 5.2 led the index at 65, and OpenAI GPT-6.2 Astra scored 62. Google said it ranks ahead of comparable models and lands just behind larger, slower models. On the Vals Index, it ranked second with 70.9 percent, and it reached first place on both the Legal Research Bench and the Tax Agent Bench.
Where Gemini 4 Neon leads on coding tests
Coding is where the model posted its strongest results. The model ranked first overall on the Artificial Analysis Coding Agent Index with a score of 58.6, and it scored 65.2 percent on Terminal-Bench 4.1 and 46.2 percent on FrontierCode 2.0. On SWE-sweep, a test of software engineering work, the model scored 12.8 percent on the shared mini-SWE harness and 39.0 percent inside Google Antigravity, the company development environment. Google said it completes most tasks about three times faster than Gemini 4 Carbon at about 60 percent lower cost.
The model also did well on tests of business and office work. It scored 56.3 percent on Zapier AutomationBench, which tests automation of real tasks, and it finished first overall there. On Andon Labs Vending-Bench 2, a long simulation of running a business, It scored 16,843 dollars. It scored 73.5 percent on Chartography, a test of reading dense professional charts. On the DualEntry Accounting AI Benchmark, the model scored 91.1 percent of accounting tasks right, tying Gemini 4 Argon for the highest score among 29 leading AI models tested.
Gemini 4 Neon results in science and math
On scientific and mathematical reasoning, the model was best in its class on three tests, according to Google. The model scored 70.5 percent on Terminal-Bench Science, 84.0 percent on RiemannBench and 71.3 percent on HLE-Diamond. It also scored 75.8 percent on Agents Last Exam version 1.1, a test of agent work. Google said it recorded its lowest fabrication rate yet in internal Antigravity testing, part of ongoing work to curb made up answers. The model card notes that Gemini 4 Neon can still produce hallucinations and that Google is continuing to improve jailbreak resistance. The model card says the full safety and security report will follow in the coming days.
Gemini 4 Neon has four selectable thinking levels. Google said the default high setting cuts cost per task by 32 percent compared with Carbon. The model has a 1 million token context window and is priced at 3 dollars per million input tokens, with cached input at 15 cents per million, and 15 dollars per million output tokens. Google said the model can synthesize text, code, images, video and other inputs in a single context window and turn them into structured deliverables.
Where Gemini 4 Neon is available now
Access opened on 18 November through the Gemini API in Google AI Studio and Android Studio, through Gemini for business, and for paid Google AI Pro and Ultra subscribers in the Gemini app and in Google Antigravity. The model is also available through GitHub Copilot. Google said Gemini 4 Neon debuted second globally with a score of 1336 on the new Agent Arena Index, based on more than 4,200 real world agent sessions. Independent checks published after launch broadly matched Google account of the model, with one review finding that Gemini 4 Neon wins most head to head benchmark rows against Claude Opus 5.5 but loses Terminal-Bench 4.1 by 0.6 points and FrontierCode 2.0 by 1.1 points.

