Google released Gemini 4 Neon on Nov. 18, a mid-tier model in its Gemini 4 family priced at $3 per million input tokens and $15 per million output tokens, according to Google’s announcement and reports from TestingCatalog and CASRAI. Google says the model delivers frontier-level performance for software engineering, knowledge work and science at a lower cost per task than its larger Carbon model.
For teams that pay for AI by the token, the main change is a new middle option: Neon sits between Carbon, built for the hardest long-horizon projects, and Hydrogen, built for fast everyday work, TestingCatalog reported.
Cached input costs $0.15 per million tokens, which Google describes as a 95% discount, and the model has a 1 million-token context window. It accepts spreadsheets, PDFs, charts, audio and video, according to TestingCatalog.
Google says that at its default high thinking level, Neon costs 32% less per task than Carbon. The model has four thinking levels so teams can trade intelligence for token spend. CASRAI noted that Carbon is priced at $4 and $20 per million tokens once promotional pricing ends on Nov. 30, and advised research offices to compare cost per task rather than list price alone.
In Google Antigravity, the company’s coding platform, Neon can act as an orchestrator that directs parallel Hydrogen subagents, Google said. GitHub also added the model to Copilot the same day, CASRAI reported, with a gradual rollout. Administrators of Copilot Business and Enterprise plans control access through the model policy in Copilot settings.
Most of these figures come from Google’s own materials, and CASRAI advised treating science and math scores as vendor-reported until independent evaluators publish results. Google also said the model had its lowest fabrication rate yet in internal Antigravity testing.
Neon is rolling out in the paid tier of the Gemini API in Google AI Studio and Android Studio. It is also available in Gemini for business through the API and model dropdown, in Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers, where it powers Tasks and deeper analysis. Eligible members of the Fairwind Program will start getting a cyber defense version over the next few days.
Google’s model card says Neon may still exhibit limitations such as hallucinations and that the company is continuing to improve resistance to jailbreaks. The card describes its safety results as automated evaluations, not human evaluation or red teaming. CASRAI noted the model’s knowledge cutoff is March 2026, which can matter for literature-review tasks.
The full Safety and Security Report for Gemini 4 Neon is due in the coming days, according to the model card.
Around Prague, based on reporting by blog.google, TestingCatalog AI News, CASRAI