Google ships Gemini 4 model family, Genie 4 API and more
- Source
- x.com
- Date
We’ve been busy shipping 🚢 Here’s a recap of the last couple of weeks: — Gemini 4 Neon, our newest model that brings frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows — Gemini 4 Hydrogen, our model for speed and efficiency in everyday coding and agentic workflows — Gemini 4 Carbon, our model for your hardest long-horizon projects and deep research — Genie 4, @GoogleDeepMind’s most realistic interactive world model featuring real-time scene editing and adaptive native audio, now available in Project Genie (with a new creation agent and gallery) and for the first time in the Gemini API and @GoogleAIStudio — Gemini 3.5 Transcribe, now with native voice editing that lets developers build experiences where users can edit, rephrase, and translate existing text in a single step using just their voice, via the Gemini API in @GoogleAIStudio — Gemini Robotics 2.5, @GoogleDeepMind’s most capable vision-language-action (VLA) model to date — AlphaGenome 2, @GoogleDeepMind’s new AI model that zooms in on specific cells to help scientists understand what drives disease — Stitch Loop, a new…
Context
This recap from Google AI collects launches that were announced one by one over roughly two weeks, and the structure it shows is a Gemini 4 lineup split by kind of work. Gemini 4 Carbon from Google DeepMind, which opened to developers and subscribers on November 10, is aimed at the longest and hardest jobs, such as multi-step debugging and open-ended research. Gemini 4 Hydrogen, released November 12, is the fast, cheaper model for routine coding and tasks. Gemini 4 Neon, released November 18, sits between them; Google's Neon announcement describes it as balancing depth and efficiency.
Google expects developers to combine these models rather than pick one. Its Hydrogen announcement describes teams using a frontier model for design decisions and Hydrogen for writing code, running tests and checking changes, and it introduced Orchestrator, a feature of Google's Antigravity coding tool that switches between Carbon and Hydrogen within one session. Until the end of November, per- prices do not follow that ranking. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, and Neon $3 and $15, while Carbon is listed at $2 and $10. That is an introductory price: Google's Carbon announcement says it rises to $4 and $20 after November 30, which puts Carbon above Neon. The same announcement says Carbon uses more tokens than Gemini 4 Argon to explore and verify its work, so a Carbon job can cost more than its per-token price suggests.
Argon is the one gap in the picture. The recap leaves it out, although Google's Carbon announcement still called Argon the right choice for most software engineering and knowledge work, and neither the recap nor the Neon announcement says how Neon relates to it.
The other new item for developers is Genie 4, a DeepMind world model that generates an explorable scene in real time from a text prompt or image and lets the user change it while moving through it. It arrived in Project Genie, Google's world-building app for AI Ultra subscribers, on November 18 and in the Gemini API a day later, where developers can generate 720p worlds with synchronized sound and steer them with keyboard controls and text prompts. The remaining entries, a voice-editing feature for Gemini 3.5 Transcribe, Gemini Robotics 2.5, AlphaGenome 2 and the Stitch Loop design experiment, get one line each, and the post gives no benchmarks, prices or limits for them.
Checking sign-in…
Loading comments…




