Gemini 4 Carbon: Google's Top Model Is Cloud Only, No Weights

Google DeepMind announced Gemini 4 Carbon on 28 October 2026 and made it generally available on 10 November 2026 (Google DeepMind, 2026). Google calls it its strongest model, built for cybersecurity and long-horizon agentic work. Short answer for local users: you cannot run it. Gemini 4 Carbon is a cloud-only model, with no published weights and no published parameter count, so it cannot run on any Mac, including a 512GB Mac Studio.

TL;DR: Gemini 4 Carbon ranks #1 on the Artificial Analysis Intelligence Index v5.0 at 59 and costs $2 per million input tokens and $10 per million output tokens at its introductory price, the same as Gemini 4 Argon. It runs only on Google's servers: the API in Google AI Studio and Gemini for business, and for Google AI Ultra subscribers in Antigravity, the Gemini app and AI Mode in Search. If you want a Google model on your own machine, the answer is Gemma 4.

What did Google announce?

Two posts by Koray Kavukcuoglu, SVP of Google DeepMind: the announcement on 28 October and the general availability post on 10 November. Between the two, Carbon was open only to a group of cyber defenders through Google's Fairwind Program.

The three security scores cover finding (SEC-bench Pro), exploiting (ExploitGym) and patching (CWE-bench) vulnerabilities. Google says they were evaluated without guardrails. On WebDev Arena, Carbon gained 89 Elo points to reach #4. Cached input tokens are priced at 95% off the input price.

Access today, per the availability post:

• Developers: the Gemini API paid tier in Google AI Studio and Android Studio.
• Enterprises: Gemini for business, through the API or the model menu, plus Antigravity and CodeMender.
• Consumers: Google AI Ultra subscribers, in the Gemini app, AI Mode in Search and Google Antigravity.

How does Carbon relate to Gemini 4 Argon?

Gemini 4 Argon came first, on 13 October 2026, about two weeks before Carbon was announced. It launched at the same introductory price of $2 per million input tokens and $10 per million output tokens, and Google says it ranked #1 on the Vals Index.

Google positions the two side by side. Argon "remains the ideal choice for the majority of software engineering and knowledge work", while Carbon is for the hardest tasks and uses more tokens than Argon to reason and verify its work. So the same price per token does not mean the same bill per task. Argon is also cloud only: neither post mentions weights for either model.

Can a 512GB Mac Studio run Gemini 4 Carbon?

No. There is nothing to download: Google has published no weights for Carbon, and the Hugging Face API lists no Gemini 4 repository under Google's account as of 12 November 2026. Google has not given a parameter count either, so the model cannot be sized against any machine. A Mac Studio with 512GB is the largest single computer we track, and it is still the wrong question for this model. Carbon runs only on Google's servers.

ModelFit does not add a model page or a dataset entry for cloud-only models. That will only change if Google publishes weights that llama.cpp or mlx-lm can load.

What can a Mac run locally instead?

Google's open-weight family is Gemma. Gemma 4 is far smaller than any frontier model, but it runs on ordinary hardware:

• Gemma 4 31B, the dense flagship, loads in about 20 GB at Q4, so a 32GB Mac runs it.
• Gemma 4 26B-A4B, the mixture-of-experts version, loads in about 16 GB at Q4 and is the faster of the two.

Our Gemma 4 GPU requirements guide lists every size in GB. If you own a 512GB Mac Studio and want the biggest model it can hold, our best LLMs for the M5 Ultra 512GB ranks the open-weight models that fit, from MiniMax M3.1 Flash to Llama 4 Maverick. None of them matches Carbon's benchmark scores. They run offline, privately, and without a per-token bill. For a pick matched to your own machine, use the ModelFit wizard or npx @modelfit/cli@latest.

For another big-lab model with no weights, see Qwen4-Max Preview.

FAQ

Can you run Gemini 4 Carbon locally?

No. Gemini 4 Carbon is cloud only. Google has published no weights and no parameter count, so it cannot run on a Mac or a PC, whatever the memory.

How much does Gemini 4 Carbon cost?

$2 per million input tokens and $10 per million output tokens at the introductory price, with cached input 95% off. After the introductory period, Google lists $4 and $20 per million.

Where can I use Gemini 4 Carbon?

Through the Gemini API in Google AI Studio and Gemini for business, and for Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Google Antigravity.

What is the difference between Gemini 4 Carbon and Gemini 4 Argon?

Argon launched on 13 October 2026 and is Google's choice for most coding and knowledge work. Carbon followed two weeks later for the hardest long-horizon and security tasks. Both have the same introductory price per token, but Carbon uses more tokens per task.

What is the best Google model I can run locally?

Gemma 4. The 31B dense model loads in about 20 GB at Q4 and the 26B-A4B mixture-of-experts model in about 16 GB.

Sources

• Google DeepMind, Gemini 4 Carbon: our next frontier model, 28 October 2026: https://deepmind.google/blog/gemini-4-carbon-our-next-frontier-model/
• Google DeepMind, Gemini 4 Carbon is now available, 10 November 2026: https://deepmind.google/blog/gemini-4-carbon-is-now-available/
• Google, Introducing Gemini 4 Argon, 13 October 2026: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/
• Hugging Face API, Google models matching "gemini-4", read 12 November 2026: https://huggingface.co/api/models?author=google&search=gemini-4