Google releases Gemini 4 Hydrogen, replacing Gemini 3.8 Flash

Google on 12 November 2026 released Gemini 4 Hydrogen, dropping the Flash name and replacing Gemini 3.8 Flash across its products. Access opened at once. Argon and Carbon rollouts began with cybersecurity researchers and, after general availability, stayed limited to the most expensive AI subscription.

Google says that among thinking levels low, medium, high, and new xhigh, high runs at 2.5 times the speed of Flash 3.8, and its analysis finds fabricated claims reduced more than five times. On Terminal-Bench, Google says Hydrogen more than triples 3.8 Flash. The New Stack says the bench versions cited are 4.0 and 4.1, and key Carbon and routing figures are unpublished.

Index scores from Google: Hydrogen 51, Argon 50 at default high; Hydrogen 54 at xhigh, also versus Argon’s 50.

Access, agents, and Copilot

In Antigravity, Orchestrator moves work between Carbon for architecture and debugging and Hydrogen for code changes and testing. Google says this stays close to using Carbon alone, at 24% less quota and 19% faster completion. Input tokens are listed at $1.50 per million and output tokens at $7.50 per million, unchanged over a one-million-token window. It is in AI Studio, the Gemini API, the Gemini app, and Google Search.

GitHub on 13 November enabled Hydrogen in Copilot for Pro, Pro+, Max, Business, and Enterprise at provider list pricing under usage-based billing, citing stronger instruction following than Gemini 3.8 Flash and still naming it Google’s latest Flash model. Rollout is gradual. Admins for Business and Enterprise use the model policy in Copilot settings; new models enable unless the global default or this model is off. The picker is in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the Copilot SDK, the coding agent, the Copilot app, github.com, and GitHub Mobile on iOS and Android.