OMM Newsroom
“Google” announced its latest artificial intelligence model, “Gemini 4 Carbon”, on Wednesday, claiming advances in coding, autonomous agent tasks and cybersecurity defense. “Anadolu Agency” reported the announcement.
Unveiled two weeks after “Gemini 4 Argon”, Carbon is the company’s strongest frontier model, with state-of-the-art cybersecurity and long-horizon capabilities, “Google” said in a blog post.
The company attributed the rapid release to researchers and AI models “working hand in hand to advance capabilities faster”.
“Google” said Carbon ranked first, with a score of 59, on the updated version 5.0 of the Artificial Analysis Intelligence Index.
Based on user feedback from the previous release, the company said it had improved the model in full-stack web development, scientific tasks and multimodal understanding. It claimed an 89-point rise to fourth place on the WebDev Arena leaderboard.
In cybersecurity benchmarks, “Google” said Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71% on CWE-bench for patching, giving it what the company called a “firm lead”.
“On internal benchmarks, Carbon drives a strong uplift in generating secure code, and in live coding environments with our own teams it shows reduced misalignment and fabrication rates,” the company said.
On safety and alignment, “Google” said internal testing showed Carbon cuts unsupported claims by approximately half compared with Argon. The company said the model states methodological caveats directly instead of fabricating missing evidence.
“Google” also said Carbon retrieves relevant sources and highlights discrepancies when faced with ambiguity. During multi-step workflows, it reduces task overreach by 26% and under-service by 29%, the company claimed.
Cybersecurity defenders will be the first to receive access to “Gemini 4 Carbon” this week, before the model is rolled out more broadly to developers, businesses and consumers, “Google” said.
The company plans introductory pricing of $2 per 1 million input tokens and $10 per 1 million output tokens. Cached input tokens will be discounted by 95%. Standard pricing will later rise to $4 per 1 million input tokens and $20 per 1 million output tokens.
