Google DeepMind released its latest frontier artificial intelligence model, Gemini 4 Carbon, on October 28, 2026. The company states that Carbon is its most capable model to date, excelling in complex reasoning, cybersecurity defense, and long-horizon tasks. Independent evaluations place Gemini 4 Carbon at the top of key benchmarks, including the Artificial Analysis Intelligence Index v5.0, where it achieved a score of 59. This positions Carbon ahead of other leading models such as Claude Opus 5.5 and GPT-6 Astra.
Gemini 4 Carbon demonstrates significant advancements in cybersecurity, showing particular strength in vulnerability detection, exploitation, and patching. Google reports that the model achieved an 88.2% success rate on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. The model also shows improved performance in coding and agentic tasks, with internal teams observing it running for days on complex engineering workflows, such as diagnosing and fixing bugs in extensive codebases. Google's Chrome team reported that Carbon outperformed another major lab's model by more than tenfold in generating high-quality patches.
The release of Gemini 4 Carbon follows closely on the heels of its predecessor, Gemini 4 Argon, which was introduced just two weeks prior. This rapid development cycle highlights the accelerated pace of progress within leading AI research labs. Google attributes this speed to its researchers and models working collaboratively to advance capabilities. The company is also emphasizing strengthened safeguards and alignment in Carbon. Internal testing indicates that Carbon exhibits reduced misalignment and fabrication rates compared to Argon, and it makes approximately half as many unsupported claims. Carbon also achieved a 0.0% attack success rate on a prompt injection benchmark, compared to Argon's 0.7%.
Access to Gemini 4 Carbon is initially restricted. Google is making the model available to a select group of cybersecurity defenders this week through its Fairwind Program. A broader release to developers, businesses, and consumers is planned after further feedback and iteration on the model's guardrails. Google is also participating in the U.S. government's voluntary pre-release model testing process.
Gemini 4 Carbon will launch with introductory pricing identical to Argon: $2 per million input tokens and $10 per million output tokens. Cached input tokens will receive a 95% discount. Google has not yet specified an end date for this introductory pricing period. Standard pricing will reportedly increase to $4 per million input tokens and $20 per million output tokens.