Gemini 4 Carbon: Our next frontier model
2026-10-29 · Google DeepMind
Gemini 4 Carbon: Our Next Frontier Model
Google DeepMind has announced its next frontier model, Gemini 4 Carbon, described as their strongest model featuring state-of-the-art cybersecurity and long-horizon capabilities. Carbon ranks #1 on the recently updated v5.0 of the Artificial Analysis Intelligence Index with a score of 59.
Improving Coding and Reasoning for Long-Horizon Tasks
Designed to tackle much harder, long-horizon engineering problems, Carbon is built for sustained agentic workflows. Internally, teams have observed Carbon running for days to persist through dead ends, tackling complex, multi-step engineering tasks such as root-causing and patching subtle bugs across large codebases and low-level systems.
Google focused on improving model performance across various domains, including coding, agentic tasks, scientific tasks, full-stack web development, and multimodal understanding. With improved output quality in web development, Carbon jumped 89 Elo points to rank #4 (1768) on WebDev Arena.
State-of-the-Art Cybersecurity Performance
Gemini 4 Carbon delivers a significant leap in cybersecurity capabilities, achieving state-of-the-art performance across the defense lifecycle: vulnerability detection, validation, and remediation.
- Finding real vulnerabilities: Carbon achieves 88.2% on SEC-bench Pro, autonomously uncovering complex flaws in real-world software.
- Validating risk through exploitation: It sets a new state-of-the-art on ExploitGym (54.8%), helping defenders prioritize which vulnerabilities are exploitable.
- Patching at the root cause: Carbon reaches 71.0% on Collinear’s CWE-bench, turning validated findings directly into working code fixes.
Internal benchmarks show Carbon achieves more than a 2x uplift in success rate for generating secure code over its predecessor, Argon. The Chrome team reported that Carbon performed over 10x better than another lab's large frontier model when generating high-quality patches. Additionally, Wiz, Google's Cloud security platform, observed Carbon leading on benchmarks for web penetration testing.
Advancing Frontier Model Safety & Security
Carbon builds on Argon’s multi-layered safety and security stack, combining model training, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring. With Carbon, Google challenged itself to go further on alignment, training the model to respond appropriately to complex or high-pressure tasks:
- Reducing shortcuts and loophole exploitation: Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts dropping to 0.4%, all while improving overall accuracy. High-severity misalignment alerts dropped by almost 80% in live internal coding environments.
- Greater transparency and honesty: Internal testing shows Carbon cuts unsupported claims by approximately 2x over Argon. It puts methodological caveats directly into deliverables or withholds blocked actions with a clear list of requirements, rather than fabricating evidence.
- Improved collaboration: Rather than guessing, Carbon retrieves relevant sources and surfaces discrepancies to the user before acting, ensuring a much better collaborative experience.
Rollout Strategy and Pricing
Google is taking a deliberate approach to rollout, actively engaged in the U.S. government’s voluntary process for pre-release model testing. Starting this week, Carbon will be available to a subset of cyber defenders, gradually phasing access to additional trusted partners through the Fairwind Program. Google will continue to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers.
Gemini 4 Carbon will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at a 95% discount.