On November 10, 2026, Google made Gemini 4 Carbon broadly available. Depending on who you are, you can reach it through the Gemini app, AI Mode in Google Search, Google AI Studio, Google Antigravity, or Gemini for business. Google calls it its best model for cyber defense and for long, multi-step “long-horizon” work.
The short verdict is that most Gemini users can ignore this one for now. Carbon is a heavy-duty model for developers, security teams, and researchers. Google itself says Gemini 4 Argon remains the ideal choice for the majority of software engineering and knowledge work. And those “agent swarm” headlines? They describe one preview feature called /swarm, and it lives inside a developer tool.
What Happened With Gemini 4 Carbon
Carbon reached the public in two steps:
- October 28, 2026: Google introduced Carbon to vetted cyber-defense partners through its Fairwind Program. Nobody else could use it yet.
- November 10, 2026: The broad rollout opened Carbon to developers, enterprises, scientists, and consumers. For consumers, that means Google AI Ultra subscribers.
What “agent swarm coordination” actually means
“Agent swarm coordination” is headline shorthand. Google doesn’t use that phrase as a product name. The real feature is /swarm, a preview command in Google Antigravity. Antigravity is Google’s agent-focused coding tool.
Think of it like a team tackling one big job together. According to Google, Carbon takes the reasoning, coordination, and auditing roles, and it’s paired with faster, lighter models that work across parallel, isolated workspaces. Google says adversarial verifiers test every change. It hasn’t published the exact mechanics of how work gets divided between the models.
The company hasn’t published enough implementation detail for outsiders to assess how reliable or configurable that verification actually is. Treat any specific claims about how well it catches errors as unconfirmed until Google says more.
Who can use Carbon, and what it costs
| Who you are | Where you get Carbon | Cost |
|---|---|---|
| Google AI Ultra subscriber | Gemini app, AI Mode in Search, and Antigravity, where /swarm is available as a preview | Included with Ultra, though Google hasn’t detailed usage limits |
| Developer | Paid tier in Google AI Studio and Android Studio (Ultra users can also try it in Antigravity and AI Studio) | Introductory API price of $2 per million input tokens and $10 per million output tokens, the same as Argon. After November 30, 2026, Google says the rate becomes $4 per million input tokens and $20 per million output tokens |
| Business | Gemini for business, through the API or the model drop-down, plus Antigravity and CodeMender | Your Gemini for business plan |
| Researcher | Apply for Argon and Carbon through Antigravity’s highest tier | Free for six months, subject to eligibility, capacity, and selection |
Google labels the API pricing “introductory” and set the November 30, 2026 cutoff itself, so the increase to $4/$20 per million tokens is Google’s stated plan, not speculation. Google notes the pricing applies only to the Carbon model intended for public access.
How Gemini 4 Carbon Stacks Up
Every claim below comes from Google’s November 10 announcement. We haven’t tested any of them ourselves, and there are no independent results confirming Google’s claims yet.
| Claim | What Google says |
|---|---|
| CursorBench 4.0, WANDR, AutomationBench | Ranked #1, at more than 2X lower cost per task than the next-best external model |
| SEC-bench Pro, ExploitGym, CWE-bench v1 | Leads these security benchmarks, according to results shared through the Fairwind Program |
| Agentic honesty benchmark | 92%, compared with 70% for Argon |
What about rivals? VentureBeat reports that Carbon’s API pricing undercuts Anthropic’s Claude Opus 5.5 and Claude Fable 5.2. That comparison is based on the introductory rate, which Google says doubles after November 30. The Verge covered the security angle at the October launch.
That’s a strong showing on paper. But it’s Google grading Google’s homework (and every AI company’s launch chart somehow puts its own bar on top).
One trade-off is easy to miss. Google says Carbon uses more tokens than Argon, so the same per-token price can still add up to a bigger bill per task. Google frames its cost advantage per task, not per token.
The Reaction
Public reaction is still forming, and we haven’t found independent tests we can verify and link yet.
Our Take
Security teams and developers with multi-day engineering problems are the people Carbon is built for. If you already use Antigravity, try /swarm on a big, well-defined job. Just remember it’s a preview, and Google hasn’t detailed how /swarm usage is billed.
Google AI Ultra subscribers get access to Carbon as part of their plan, though Google hasn’t detailed usage limits. Save it for hard research or tricky analysis. Stick with Argon for most everyday work, as Google itself advises.
Free and Pro Gemini users aren’t mentioned in Google’s rollout, which lists Ultra as the consumer plan with access. You aren’t missing much for email drafts and trip planning.
ChatGPT and Claude users shouldn’t switch over benchmarks alone. The only figures so far are Google’s own.
Researchers can apply for Argon and Carbon through Antigravity’s highest tier, free for six months. Access is subject to eligibility, capacity, and selection, so it isn’t guaranteed.
The one practical next step: if you pay for Ultra, open the model picker in the Gemini app. Check whether Carbon is listed for your account.
Wrapping Up
Gemini 4 Carbon is a specialist tool, and the “agent swarm” buzz describes the /swarm preview in Antigravity. It’s a real feature, but it’s aimed at developers. Use Carbon if you hunt security holes or untangle huge codebases. Stick with Argon, or whatever chatbot you already like, if you mostly use AI for everyday writing and questions.