Google today revealed its all-new Gemini 4 Carbon, making it available to developers, enterprises, scientists and consumers. According to Google, Carbon offers frontier-level performance in cybersecurity and long-horizon agentic workflows across complex knowledge work, software engineering and scientific reasoning. Following the recent launch of Argon, Carbon pushes the frontier even further.

Google adds that Carbon is built for tasks that test the limits of a model's intelligence, such as complex, multi-step coding and debugging, as well as open-ended scientific research and discovery. It uses more tokens than Argon to explore problems thoroughly, reason deeply and run extended verification loops.

Across independent benchmarks designed to test model performance in real-world workflows, Carbon combines frontier-level accuracy with practical production capabilities.

To ensure its capabilities can tackle the most demanding tasks, Google is working on larger and more complex projects, including a preview of Swarm in Google Antigravity. The feature turns a single prompt into a coordinated team of agents that users can track in real time. Users will be able to assign Carbon roles that demand deeper reasoning, such as architectural design, coordination and auditing, while pairing it with faster, lighter models across parallel, isolated workspaces and adversarial verifiers that test every change.

Improved agentic coding capabilities

With its improved capabilities, Carbon performs particularly well on engineering tasks that require multi-agent collaboration, such as multi-package migrations, cross-subsystem root-cause investigations and complex architectural redesigns.

For web and app development, it runs build-and-verify loops, recovers from errors and checks both its code and the rendered user interface to produce complete, working applications. Stronger visual design, spatial reasoning and attention to environmental detail also extend to 3D game development and creative coding.

Cybersecurity capabilities

Google says Gemini 4 Carbon for Cyber brings its strongest cybersecurity capabilities yet, helping security teams identify, exploit and patch software vulnerabilities. Available through its Fairwind Program, the model has demonstrated improvements across key security benchmarks, including vulnerability detection, exploitation and patching.

Through Wiz's Scan for Good programme, Carbon identified zero-day vulnerabilities in the public digital infrastructure of two major SaaS providers in under two hours, significantly reducing the time required for vulnerability research. The affected organisations worked with the team to address the flaws, potentially protecting tens of thousands of companies.

Stronger scientific and mathematical reasoning

Carbon also demonstrates improved reasoning capabilities, with Google saying it is now the leading model on Axiom's benchmark for performance in transactional science. It reportedly helped Axiom identify previously undiscovered mechanisms of toxicity. Axiom is now moving its production scientific analyses and agents to both Carbon and Argon to support the development of safer drugs.

Beyond cybersecurity, Gemini 4 Carbon is also being deployed across scientific research, with Google highlighting applications in formal code verification, laboratory research and protein structure prediction.

In one major development, researchers used Carbon alongside other models to support AlphaProof Code, an agentic framework designed for formal verification. The system helped mathematically prove key security properties of the cryptographic protocol underlying Signal by generating millions of lines of code in Lean, a programming language used for mathematical proofs. According to Google, a task that could historically take expert teams years was completed in weeks, highlighting the potential of AI-assisted formal verification to improve software security.

In laboratory research, Carbon helped scientists identify anomalies in enzyme assay data, including incomplete automated mixing and sample carryover in research instruments. These findings helped improve experimental reproducibility, refine chromatography analysis and accelerate the development of new assays.

The model has also demonstrated improvements in protein structure prediction. Working with multiple AI agents, Carbon autonomously identified evolutionary relatives of understudied proteins across large, previously unannotated biological databases.

Expanded access for researchers

Google is also expanding access to Carbon through its existing research programme on Google Antigravity. Initially, the company opened applications to provide 10,000 scientists with six months of free access to Argon. Thousands of researchers have already joined the programme. Google is now extending access to Carbon for existing and new participants, with plans to continue offering its latest frontier models as they become available. Applications remain open, with researchers being onboarded weekly, subject to eligibility, capacity and selection.

Availability

Google is rolling out Gemini 4 Carbon across its products for developers, enterprises and consumers. Developers can access it through the paid Gemini API tier in AI Studio and Android Studio, while Google AI Ultra subscribers can also try it through Google Antigravity. Enterprise users can access the model through Gemini for business, its API, Google Antigravity and CodeMender.

For everyday users, Carbon is available through the Gemini app and AI Mode in Search for Google AI Ultra subscribers. Scientists and researchers can apply for six months of free access to Argon and Carbon through Google Antigravity, subject to eligibility and selection. Cybersecurity professionals can also apply to access Carbon for Cyber through Google's Fairwind Program.