Image: Google
Google has released Gemini 4 Neon, a new model that sits between the top-end Gemini 4 Carbon and the cheaper Gemini 4 Hydrogen. It is the model that turned up in Google Cloud billing listings a few days ago. Google pitches it for complex coding, knowledge work and science, with an emphasis on finishing tasks quickly.
Gemini 4 Neon is rolling out now.
— Google DeepMind (@GoogleDeepMind) November 18, 2026
Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science. pic.twitter.com/qfOLW06FOm
Neon costs $3 per million input tokens, $0.15 for cached input and $15 per million output tokens, with a 1 million-token context window. It is rolling out in the Gemini API through Google AI Studio, in Gemini for business, in Google Antigravity for paid AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. Google's Logan Kilpatrick called it "the right balance of intelligence, speed, and usability."
Introducing Gemini 4 Neon! Pushing the frontier across domains and punching above its model class in a meaningful way. To me, Neon feels like the right balance of intelligence, speed, and usability. Available now across @GoogleAIStudio, the API, and more! pic.twitter.com/ANqCSTEWcS
— Logan Kilpatrick (@OfficialLoganK) November 18, 2026

Image: Google
The benchmark table Google shared puts Neon against Claude Opus 5.5 and OpenAI's GPT-6.2 Sol, and Neon leads on most rows. It scores 61 on the Artificial Analysis Intelligence Index, against 59 for Opus 5.5 and 57 for Sol, and 39.0% on the SWE-sweep coding test in Google's own Antigravity harness, against about 10% for the others. It doesn't win everywhere. Opus 5.5 edges it on Terminal-bench 4.1 and FrontierCode 2.0, and Sol is ahead on chart reading. The table also leaves out the biggest rival models, Claude Fable 5.2 and GPT-6.2 Astra. Google says Neon ranks third overall on the Artificial Analysis index behind "larger, slower models," which is a fairer description of where it sits. The SWE-sweep figure also shows how much the testing setup matters. Neon scores 12.8% on that benchmark's neutral harness and 39.0% in Antigravity.
Cost is the other half of the pitch. At its default "high" thinking level, Google says Neon costs 32% less per task than Carbon, and 50% to 80% less than top competing models on two third-party leaderboards. Glean, a customer quoted in the announcement, said Neon matched or beat Carbon in its tests at three times the speed and 60% of the cost.

Image: Google
Google's demos lean on Antigravity's agent tools. In one, Neon used a team of agents to rebuild Alan Turing's 1936 study at King's College, Cambridge, in interactive 3D from archival photos. In another, it wrote a ski simulator with more than 10,000 lines of custom graphics code. Google also says Neon can act as an orchestrator that hands work to parallel Hydrogen agents.
A security version, Gemini 4 Neon for cyber, will reach vetted defenders through Google's Fairwind Program in the coming days. On safety, Google says Neon keeps Carbon's protections while refusing fewer legitimate requests, and that it cheated on 0% of Terminal-Bench 4.0 tasks and took shortcuts on 2.45% of tasks in an internal coding test, the lowest of any Gemini 4 model. Those safety figures come from Google's internal tests, and Google chose which rivals and benchmarks to show in its comparisons. Broader independent testing is still to come.
Comments