AI Tech Flash ── 2026-11-11
Google Releases Gemini 4 Carbon Model
Google opens a model for hard reasoning; prices double Nov. 30
What changes
According to Google DeepMind, Carbon is built for complex, multi-step work. Examples include multi-step coding and debugging, and open-ended scientific research.
The company still recommends Gemini 4 Argon, a model in the same family, for most software engineering and knowledge work. It says Carbon is for tasks that test the limits of a model's intelligence.
Why it matters
In Google's announcement, Carbon ranks No. 1 on three benchmarks (standardized tests used to compare models): CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench. Google says its cost per task is more than 2X lower than the next best external model.
These figures come from Google's own announcement. According to DigitalToday, Carbon's prices double after the introductory period ends.
What it is
Gemini 4 Carbon is Google's strongest AI model. Google DeepMind says it delivers frontier performance in cybersecurity and in long-horizon agentic workflows. Agentic workflows are tasks in which an AI carries out many steps by itself.
Google's announcement says it also suits document analysis in legal and finance work, software engineering, and scientific reasoning.
Who and where
Google DeepMind built it. Its official blog post of Nov. 10, 2026 announced availability to developers, enterprises, scientists and consumers.
The API is offered in Google AI Studio and Gemini for business to paying developers and enterprises. Google AI Ultra subscribers are getting it in Google Antigravity, the Gemini app and AI Mode in Search. According to DigitalToday, Ultra costs $100 a month.
Why it was built
According to Google's announcement, Carbon was built for tasks that test the limits of a model's intelligence. Examples are complex, multi-step coding and debugging, and open-ended scientific research.
Google positions Carbon for these tasks and keeps Argon for more general work. Argon was released last month.
How it works
According to Google DeepMind, Carbon uses more tokens than Argon. A token is a unit of text that a model processes. Carbon spends them to explore thoroughly, reason deeply and run extended verification loops. DigitalToday likewise reports that it is designed to perform more verification steps before answering.
Google is also previewing /swarm in its development platform, Google Antigravity. It turns a single prompt into a team of agents that users can track in real time. Developers can assign Carbon to roles such as design and auditing, and use lighter models for parallel work. Adversarial verifiers test every change.
Key points
- Google says Carbon ranks No. 1 on CursorBench 4.0, WANDR and AutomationBench.
- API pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after Nov. 30.
- Antigravity's /swarm preview turns one prompt into a team of agents.
Conclusion
Google DeepMind has released Gemini 4 Carbon for hard reasoning tasks. The performance and cost figures come from Google's own announcement. The introductory price runs through Nov. 30, 2026 and then doubles.
Quotes and sources
“Carbon is our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning.”── Google DeepMind, 2026-11-10
“Carbon ranks #2 while being more than 6X lower in average cost per task compared to the top performing model.”── Google DeepMind, 2026-11-10
“Carbon is designed to perform more verification steps before answering.”── 디지털투데이, 2026-11-11
“The /swarm preview turns a single prompt into a team of AI agents that can be tracked in real time.”── Inven Global, 2026-11-11
Source articles
- Google DeepMind 11/10 16:03 UTC
- 디지털투데이 11/11 09:11
- Inven Global 11/11 08:28
Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.