TL;DR: Google opened Gemini 4 Carbon to paying developers, enterprises and Google AI Ultra subscribers on November 10, 2026. The Gemini API model ID is gemini-4-carbon. The introductory price is $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4/$20 from December 1. The same day, Google previewed /swarm in Google Antigravity, and DeepMind introduced AlphaProof Code, which used Carbon to generate more than six million lines of Lean proofs for key security properties of Signal’s cryptographic code. Two days later, on November 12, Google launched Gemini 4 Hydrogen, a faster model adapted from Carbon and priced at $1.50/$7.50.
Last updated: November 13, 2026. Prices are in USD, as billed on the Gemini API (the same dollar rates apply in the UK). Every fact below comes from Google’s or DeepMind’s own pages, or the official Gemini API pricing page, checked November 13, 2026. Google’s benchmark claims are quoted as Google published them; we have not reproduced those scores.
Gemini 4 Carbon is Google’s strongest Gemini 4 model for long-horizon agent work, software engineering, scientific reasoning and cybersecurity. It was already in the hands of some cyber defenders through Google’s Fairwind Program. November 10 is when paying API customers and Ultra subscribers could actually call it. For the rest of this week’s AI launches, see our AI News This Week roundup for November 9–15.
What is Gemini 4 Carbon and when did it come out?
Gemini 4 Carbon is Google DeepMind’s strongest Gemini 4 model, and Google made it available to paying developers, enterprises and Google AI Ultra subscribers on November 10, 2026. Google describes Carbon as its model with frontier performance in cybersecurity and in long-horizon agentic workflows across knowledge work, software engineering and scientific reasoning.
In the November 10 announcement, Koray Kavukcuoglu, Google DeepMind’s chief AI architect, wrote that Carbon “pushes the frontier even further” after Gemini 4 Argon. Argon, Google said, “remains the ideal choice for the majority of software engineering and knowledge work,” while Carbon is built for tasks that “test the limits of a model’s intelligence,” such as complex multi-step coding and debugging and open-ended scientific research. Carbon “uses more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops.”
Carbon is based on Argon, according to the Gemini 4 Carbon model card. It accepts text, images, audio and video, with a unified input-plus-output context window of up to 1 million tokens. The knowledge cutoff listed on that model card is March 2026. The public model ID on the Gemini API is gemini-4-carbon. Thinking depth is controlled with thinking_level set to low, medium (the default) or high.
Google said it had already released Carbon to a set of cybersecurity defenders through its Fairwind Program two weeks before the November 10 public launch.
How much does Gemini 4 Carbon cost?
On the paid Gemini API, Gemini 4 Carbon costs $2.00 per million input tokens and $10.00 per million output tokens through November 30, 2026, then $4.00 / $20.00 from December 1, 2026. That is the same introductory schedule as Gemini 4 Argon. Output pricing includes thinking tokens. Carbon is not available on the Free tier.
These figures are from the Gemini Developer API pricing page, checked November 13, 2026, and they match the footnote in Google’s November 10 post. Implicit context caching is supported; explicit caching is not. Every token the model processes is billed, including thinking tokens and, when you use grounding, internal tool system instructions and function declarations.
Gemini 4 Carbon (gemini-4-carbon), per 1M tokens (USD) |
Through November 30, 2026 | From December 1, 2026 |
|---|---|---|
| Standard input | $2.00 | $4.00 |
| Standard output (including thinking tokens) | $10.00 | $20.00 |
| Standard implicit cached input | $0.10 | $0.20 |
| Batch or Flex input | $1.00 | $2.00 |
| Batch or Flex output | $5.00 | $10.00 |
| Batch or Flex implicit cached input | $0.05 | $0.10 |
| Priority input | $3.60 | $7.20 |
| Priority output | $18.00 | $36.00 |
| Priority implicit cached input | $0.18 | $0.36 |
| Free tier | Not available | Not available |
Grounding with Google Search or Maps is billed separately after a shared monthly allowance, as listed on the same pricing page. For consumer plan prices, see our ChatGPT vs Gemini comparison and ChatGPT Plus vs Claude Pro vs Google AI Pro.
Who can use Gemini 4 Carbon today?
Paying Gemini API customers and Google AI Ultra subscribers can use Gemini 4 Carbon today; free Gemini app users and Google AI Pro subscribers do not get the same access. Google’s November 10 post and the Carbon model card spell out the channels:
- Developers: the paid Gemini API in Google AI Studio and Android Studio. Google AI Ultra subscribers can also use Carbon in Google Antigravity and AI Studio.
- Enterprises: Gemini for business (API or the model dropdown), Google Antigravity, or CodeMender.
- Everyday users: the Gemini app and AI Mode in Search, if you are a Google AI Ultra subscriber.
- Scientists and researchers: can apply for six months of free access to Argon and Carbon at Google Antigravity’s highest usage tier, subject to eligibility, capacity and selection. Google is expanding a 10,000-seat scientist program to include Carbon.
- Cybersecurity defenders: can apply to the Fairwind Program for Carbon’s specialized cyber capabilities, on a rolling basis.
Google AI Pro is enough for Hydrogen in Antigravity, but Carbon in the Gemini app, AI Mode and Antigravity is an Ultra-tier feature.
What is AlphaProof Code?
AlphaProof Code is a DeepMind agentic framework for formal verification that, with human experts in the loop, mathematically proved key security properties of the actual code behind the Signal protocol. DeepMind published the work on November 10, 2026, the same day Carbon opened more widely.
Formal verification means proving, in mathematics, that code meets a written specification, rather than only probing it with examples. DeepMind said this is the first time AI has been used to formally verify a widely deployed cryptographic protocol at this depth, and the first time the implementation of a widely deployed protocol like this has been formally verified by any method, with or without AI. Most protocol verification, DeepMind noted, stops at the design level; this work aimed at the compiled code that runs on devices.
According to the AlphaProof Code announcement, researchers first worked with a mix of models to turn Signal’s security goals into formal specifications, which human experts then checked. Agents powered by Gemini Carbon then generated the proofs — more than six million lines of Lean, a language used to write machine-checkable proofs. The agents also flagged about one percent of the codebase for a second look by a human; Lean automatically checks the rest. DeepMind said the process took four weeks, “a task that historically takes expert teams years,” and that it covered properties such as keeping messages protected before and after a device is temporarily compromised.
To check that the system could tell correct code from buggy code, the team reintroduced two previously fixed bugs into an isolated copy of Signal’s code. In both cases, DeepMind said, it produced a machine-checked proof that the buggy version was insecure. DeepMind is open-sourcing the artifacts.
Ehren Kret, Signal’s chief technology officer, said the approach “enables significant expansion” on prior formal verification and that “using AI agents to translate live code into valid Lean theorems is a major breakthrough.” DeepMind said the Signal protocol also protects conversations in Google Messages. AlphaProof Code is a research milestone, not a product you can buy today. For the broader questions around giving agents tools, see our AI agent security checklist.
What is /swarm in Google Antigravity?
/swarm is a preview feature in Google Antigravity that turns one prompt into a coordinated team of agents: a coordinator, specialized workers in isolated git worktrees, and adversarial verifiers that check changes. Google announced it on November 10, 2026, alongside Carbon.
The Antigravity blog post says a single agent works for scoped tasks but struggles when a change spans many files. With /swarm, the prompt goes to a family of agents that plan, build and verify in parallel. Workers edit in isolated git worktrees so changes do not collide. Agents share an append-only message board, sit idle at zero token cost, and wake when another agent needs input. A Swarm UI in the side pane (and /swarm board or Alt+S in the CLI) shows a Board feed, a Members roster and an interactive Graph timeline.
Agents can be pinned to different models. Google says Carbon fits coordinator, architecture, cybersecurity auditing and adversarial verification roles, while faster, lighter models handle parallel file edits. Google AI Ultra subscribers can use Carbon in the Antigravity app, CLI and IDE extensions. /swarm is rolling out in preview in the app and CLI. Google warned that even with zero-token idle waits, “parallel multi-agent systems consume quota much faster than a standard chat.”
How does Gemini 4 Carbon compare to Gemini 4 Hydrogen?
Gemini 4 Carbon is the deeper, more expensive frontier model; Gemini 4 Hydrogen, launched two days later on November 12, 2026, is the faster execution model, and DeepMind’s model card says Hydrogen is adapted from Carbon. Carbon is for the hardest long-horizon work and is limited to paid API customers and Ultra subscribers. Hydrogen is priced lower, is available much more widely, and is the model Google is putting in Search for everyone.
Google’s Hydrogen announcement says Carbon and Argon “excel at deep, long-horizon reasoning on your hardest problems,” while Hydrogen is “built for rapid execution across diverse agentic workflows.” Hydrogen costs a flat $1.50 per million input tokens and $7.50 per million output tokens across its 1 million-token context window. It has four thinking levels — low, medium, high (the default) and a new xhigh — and is available in the Gemini API and AI Studio for all tiers, in Search for all users, and in Antigravity for paid Google AI Pro and Ultra subscribers. Free Gemini app users can get Hydrogen when Google’s auto model selection picks it.
To pair the two automatically, Google added Orchestrator in Antigravity. Orchestrator switches between models such as Carbon and Hydrogen in one session. On Terminal-Bench 4.0, Google said Orchestrator reached near-Carbon task success rates while, on average, using 24% less quota and finishing 19% faster. The Antigravity Hydrogen post says Orchestrator (Carbon) and Orchestrator (Argon) are rolling out for Google AI Ultra subscribers, with Orchestrator (Argon) for paid Google AI Pro subscribers.
Argon went generally available on October 13, 2026, at the same $2/$10 introductory API price as Carbon. Our earlier Gemini 4 Argon release date post tracked that price and the Fairwind-first rollout.
| Gemini 4 Carbon | Gemini 4 Hydrogen | Gemini 4 Argon | |
|---|---|---|---|
| Google’s role for it | Strongest / deepest long-horizon model | Fast everyday coding and agent execution | Frontier model for most software engineering and knowledge work |
| Public launch | November 10, 2026 (Fairwind access about two weeks earlier) | November 12, 2026 | Generally available October 13, 2026 (announced September 30) |
| Relationship | Based on Argon (DeepMind model card) | Adapted from Carbon (DeepMind model card) | Earlier Gemini 4 frontier model |
| Standard API price (per 1M tokens) | $2 / $10 through November 30, then $4 / $20 | $1.50 / $7.50, flat | Same schedule as Carbon |
| Context window | Up to 1M input + output | 1M input; up to 1M output (64K default synchronous) | Up to 1M input + output |
| Thinking levels | low, medium (default), high |
low, medium, high (default), xhigh | Same family of controls as Carbon; see Google’s Argon docs for the current list |
| Who can use it | Paid Gemini API; Ultra (app, Search AI Mode, Antigravity); Fairwind; scientists program | API and AI Studio (all tiers); Search for all users; Antigravity for paid Pro and Ultra; free Gemini app via auto-select | Paid Gemini API and Ultra, per Google’s Argon launch |
| Headline score Google cites | Artificial Analysis Intelligence Index v5.0: 59 (Carbon model card, results as of October 2026) | 51 at default high thinking; 54 at xhigh (Google Hydrogen blog) | 50 on the same index, per both the Carbon model card and the Hydrogen blog |
Those index scores are Google’s published figures, not scores we reproduced. They also sit on slightly different write-ups (the Carbon model card vs the Hydrogen blog), so treat them as directional, not as a single official league table.
What did Google claim about Carbon’s performance?
Google says Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, at more than 2× lower cost per task than the next-best external model, and that it leads several cybersecurity benchmarks when used through Fairwind. Those claims are Google’s. They have not been independently reproduced here.
Other figures from the November 10 blog and the Carbon model card (results “as of October 2026”; cybersecurity scores are for Fairwind access):
- Artificial Analysis Intelligence Index v5.0: Carbon 59, Argon 50, GPT-6 Astra 57, Claude Opus 5.5 58.
- Terminal-Bench 4.0 (Artificial Analysis): Carbon 65.2%, Argon 57.1%.
- SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%.
- On a new Multimodal Knowledge Agent (MKA) benchmark with Artificial Analysis, Google said Carbon ranks #2 at more than 6× lower average cost per task than the top model.
- In Project Fortify, Google reported 100% recall on scanning and 95% precision on triage on an internal benchmark, and more than a 25% relative improvement on Argon’s recall rate on open-source repositories.
- Safety (Google’s figures): a 0.8% per-attempt breach rate, down from 3.1% on Argon; Gray Swan automated cyberattack query success of 2.17%, down from 9.86%; a 0% prompt-injection success rate on the Gray Swan IPI benchmark, which Google called a first, while saying robustness still needs work; and 92% on a 375-task agentic honesty benchmark, versus 70% for Argon. Trajectory Labs, Google said, found zero new universal jailbreaks across 132 hours and more than 13,000 requests.
Timeline
| Date | What happened |
|---|---|
| About two weeks before November 10, 2026 | Google says it released Carbon to a set of cybersecurity defenders through the Fairwind Program |
| November 10, 2026 | Google opens Gemini 4 Carbon to paying developers, enterprises and Google AI Ultra subscribers; previews /swarm; DeepMind publishes AlphaProof Code |
| November 12, 2026 | Google launches Gemini 4 Hydrogen and Orchestrator in Antigravity |
| November 30, 2026 | Introductory Carbon (and Argon) API prices end at the close of this date |
| December 1, 2026 | Carbon standard API price becomes $4 per million input tokens and $20 per million output tokens |
Our verdict
If you already pay for the Gemini API or Google AI Ultra and you have a hard, long-running coding, research or verification job, Gemini 4 Carbon is the new top Gemini 4 option, at the same $2/$10 introductory rate as Argon until November 30. For everyday coding, most people will want Hydrogen, which costs less, is available more widely, and can call Carbon through Orchestrator when a task needs it.
The AlphaProof Code result is the most concrete research claim in the launch: a machine-checked proof effort on real Signal code, in four weeks, with humans still writing and checking the specifications. /swarm is a preview, and Google itself flags that it will burn quota quickly. None of the benchmark numbers above are scores we reproduced; if a number matters to a buying decision, open Google’s post or the model card and compare it with an independent leaderboard when one is updated.
FAQ
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google DeepMind’s strongest Gemini 4 model, opened to paying developers, enterprises and Google AI Ultra subscribers on November 10, 2026. Google positions it for long-horizon agent work, complex software engineering, scientific reasoning and cybersecurity, and says it uses more tokens than Gemini 4 Argon to reason more deeply.
How much does Gemini 4 Carbon cost?
On the paid Gemini API, Carbon costs $2.00 per million input tokens and $10.00 per million output tokens (including thinking tokens) through November 30, 2026, then $4.00 / $20.00 from December 1. Batch and Flex are $1.00 / $5.00, then $2.00 / $10.00. It is not available on the Free tier. Prices are in USD.
Who can use Gemini 4 Carbon?
Paying Gemini API customers can call gemini-4-carbon in Google AI Studio and Android Studio. Google AI Ultra subscribers also get it in the Gemini app, AI Mode in Search, and Google Antigravity. Scientists can apply for a six-month Antigravity grant, and cyber defenders can apply to the Fairwind Program. Free Gemini app users and Google AI Pro subscribers do not get the same Carbon access.
What is AlphaProof Code?
AlphaProof Code is a DeepMind agentic formal-verification framework announced on November 10, 2026. With human experts setting and checking specifications, Carbon-powered agents generated more than six million lines of Lean proofs for key security properties of the Signal protocol’s implementation, in four weeks. DeepMind is open-sourcing the artifacts; it is not a generally available product.
What is /swarm in Google Antigravity?
/swarm is a preview in the Antigravity app and CLI that turns one prompt into a coordinator, parallel workers in isolated git worktrees, and adversarial verifiers. Agents share an append-only message board and can be pinned to different models, including Carbon. It is available to Google AI Ultra subscribers. Google warns that swarms use quota much faster than a normal chat.
How is Gemini 4 Carbon different from Gemini 4 Hydrogen?
Carbon is the deeper, higher-priced frontier model (starting at $2/$10, then $4/$20) and is limited to paid API and Ultra access. Hydrogen, launched November 12, 2026, is adapted from Carbon, costs a flat $1.50/$7.50, and is available much more widely, including in Search for all users. Antigravity’s Orchestrator can switch between them in one session.
Is Gemini 4 Carbon better than Gemini 4 Argon?
Google says Argon is still the right default for most software engineering and knowledge work, and that Carbon is for harder, longer tasks that use more tokens. On the Artificial Analysis Intelligence Index v5.0, the Carbon model card lists Carbon at 59 and Argon at 50. Those are Google’s published figures, not scores we reproduced.
Is Gemini 4 Carbon free to use?
No. Carbon is not on the Gemini API Free tier. The exception is a limited scientists-and-researchers program that offers six months of Antigravity’s highest usage tier at no charge, subject to eligibility, capacity and selection, plus Fairwind access for approved cyber defenders.
Sources
Google, Gemini 4 Carbon is now available, November 10, 2026; Google Antigravity, Gemini 4 Carbon and /swarm, November 10, 2026; DeepMind, Introducing AlphaProof Code, November 10, 2026; DeepMind, Gemini 4 Carbon model card; Gemini Developer API pricing (checked November 13, 2026); Google, Introducing Gemini 4 Hydrogen, November 12, 2026; DeepMind, Gemini 4 Hydrogen model card; Google Antigravity, Gemini 4 Hydrogen and Orchestrator, November 12, 2026.*
