What happened
Google’s Gemini API documentation lists two models under a new “Gemini 4” section: Gemini 4 Carbon and Gemini 4 Argon. The page gives their model codes as gemini-4-carbon and gemini-4-argon, and includes “Try in AI Studio” links for each, alongside links to model documentation. This establishes that Google is presenting the models in its developer catalog and directing users to try them. The page does not describe them as Gemini 4 Flash models.
Google calls Carbon its “strongest reasoning model,” built for complex software engineering, multimodal knowledge work, scientific reasoning and cybersecurity defense. It lists a context window of 1 million tokens and a maximum output of “65K / 1M tokens.” Its supported modalities are text, image, audio, video and PDF.
Argon is described as an “efficient Gemini 4 frontier reasoning model.” Google says it delivers faster responses and lower token spend per turn than Carbon across complex coding, agentic, math and science tasks. The catalog lists the same context window, maximum output and modalities as Carbon: 1 million tokens; 65K / 1M tokens; and text, image, audio, video and PDF. The page does not define the conditions behind the speed and token-spend comparison.
The wider model catalog places the two entries alongside Gemini 3 models and older model families. Elsewhere, Google describes Gemini 3.8 Flash as engineered for long-horizon software engineering, autonomous agents and complex enterprise workflows. That is a separate model listing; the page does not say that Carbon or Argon replaces it. Nor does the Gemini 4 section list a “Hydrogen” endpoint. The evidence supports the appearance of Carbon and Argon in Google’s catalog, not a Hydrogen release.
The page gives no price, rate limits, launch date or general availability statement for Carbon or Argon. It points to AI Studio for trying them, but does not spell out eligibility, geographic availability or whether API access is open to all developers. Those details matter before treating a catalog listing as a production-ready launch.
What the archive adds
New Runtime’s earlier coverage of Gemini 3.8 Flash TTS and Gemini 3.8 Flash described Google’s Gemini 3.8 product surface. This is a new event: the catalog now lists Gemini 4 Carbon and Argon with separate positioning and specifications. The earlier note is useful lineup context, but it does not establish a direct successor relationship.
Trend signal
This supports, weakly, agent economics moves to completed work: Google frames Argon partly around faster responses and lower token spend per turn on agentic tasks. Those are inputs to agent cost, not evidence of lower cost per verified completed task. The page publishes no benchmark, task-completion rate or total-workflow cost comparison, so the stronger economic claim remains untested.
Why now
The catalog positions Carbon and Argon within the Gemini 4 family while continuing to list Gemini 3 offerings. That gives builders a new pair of model endpoints to evaluate against existing choices. Google’s page does not explain the timing of the addition or connect it to a release schedule.
Why it matters
For teams building coding or research agents, the distinction Google draws between Carbon and Argon suggests two candidate operating points: a model positioned for stronger reasoning and a model positioned for faster responses and lower token spend per turn. That can guide a test plan, but it is not enough to select a production model. The catalog does not disclose prices, measured latency, task success rates or the evaluation conditions for Argon’s efficiency claims.
The matching 1-million-token context windows and listed modalities make the models appear comparable on those specifications. They do not show that either model will handle a team’s particular repository, document set or tool chain better. Builders should compare both on representative tasks, including retries, tool calls and human review—not only single-turn prompts. The Hydrogen rumor should not be used as an integration target: the listed model codes are Carbon and Argon.
Limits and uncertainty
Google’s catalog does not provide pricing, rate limits, an explicit general-availability status, a launch date, or details about access eligibility for these two listings. It gives no benchmark figures or methodology to quantify the claims about Argon’s speed and token spend relative to Carbon. The “65K / 1M tokens” maximum-output notation is not explained on the page. A listing and “Try in AI Studio” links do not, by themselves, establish that every developer can use the models through the API in production. The page also does not substantiate the Hydrogen name circulated in the rumor.
Skill to build now
Create a small evaluation set from real coding or agentic tasks your team already reviews. If you have access to both models, run the same inputs, tools and stopping conditions against Carbon and Argon; record completion quality, latency, token use and the number of human corrections. Keep the model codes in the results so that a later endpoint or alias change does not blur which version you tested.
Reader takeaway
Google’s catalog lists Gemini 4 Carbon and Argon, not Hydrogen. It positions Carbon for stronger reasoning and Argon for faster responses and lower token spend per turn, but offers no public figures here to show how those claims translate into production cost or task performance. Treat the entries as candidates for an evaluation, not as a finished model-selection decision.