Google Releases Gemini 4 Carbon, Strengthening Cybersecurity and Long-Running Agent Workflows
Two weeks after the debut of Gemini 4 Argon, Google released a new-generation frontier model, Gemini 4 Carbon, saying it achieves the current highest level in cybersecurity and long-horizon tasks, with enhancements for coding, Agent workflows, and cyber defense. The model scored 88.2% on the vulnerability discovery benchmark SEC-bench Pro, 54.8% on the exploit benchmark Exploit Gym, and 71.0% on the patch application benchmark CWE-bench. Carbon will first be made available this week to some cybersecurity defenders, then expanded to more trusted partners, and only afterward released to developers, enterprises, and consumers; during the trial period it follows Argon pricing at $2 per million input tokens and $10 per million output tokens, adjusted to $4 and $20 after the trial period.
Why read this
SelectedAI score 77
The original text lists Carbon's specific scores on benchmarks such as vulnerability discovery and patch application, as well as its phased rollout, which can be used to assess its positioning in security capabilities.
Topics
Body · AI translation
Download the app
+1 Google releases Gemini 4 Carbon, focusing on cybersecurity and long-horizon work
The Paper reporter Ji Jingjie 2026-10-29 11:54 Source: The Paper ∙ Future 2% > Listen to full text Font size
Just two weeks after the debut of Gemini 4 Argon, Google on October 28 local time released another new-generation frontier model, Gemini 4 Carbon. Google said in an official blog post that the model has reached the current highest level in cybersecurity and long-horizon tasks, and has been specifically strengthened around three major scenarios: coding, Agent workflows and cyber defense.
The release approach for Carbon continues Argon's cautious strategy: this week the model is first opened to a subset of cybersecurity defenders, then gradually expanded to more trusted partners, and only after collecting feedback and iterating on safeguards will it be rolled out to developers, enterprises and consumers through product interfaces.
Google disclosed that Gemini 4 Carbon performs excellently across the entire computer security defense cycle of "vulnerability discovery, verification and remediation." It scored 88.2% on the vulnerability discovery benchmark SEC-bench Pro, 54.8% on the exploit benchmark Exploit Gym, and 71.0% on the patch application benchmark CWE-bench, leading on all three metrics.
In Google's own tests, Carbon's success rate in generating secure code more than doubled compared with the previous-generation model Argon. The Chrome team reported that in generating high-quality patches, Carbon performed more than 10 times better than "a large frontier model from another lab."
Carbon's engineering goals point clearly to long-horizon, multi-step, continuously running Agent workflows. Google said its internal teams have observed Carbon "running continuously for several days" to tackle complex engineering problems, such as locating and patching hidden defects in large codebases and low-level systems.
In front-end and full-stack development scenarios, Carbon ranked fourth on the WebDevArena leaderboard with 1,768 points. Google said this improvement comes from simultaneous gains in output quality and the quality of web development artifacts, while the model has also been specifically optimized for coding, Agent tasks, scientific tasks and multimodal understanding.
On behavioral reliability, Carbon has also made engineering adjustments, reducing unsupported claims by about half, excessive overreach by 26%, and under-service behavior by 29%, with the goal of neither overstepping nor falling short in executing multi-step plans.
On pricing, Carbon currently follows Argon's pricing during the trial period: $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input tokens. After the trial period, standard pricing will be adjusted to $4 per million input tokens and $20 per million output tokens.
Responsible editor: Huan Yanhong Image editor: Shen Ke Proofreader: Zhang Yan The Paper news tip: 021-962866 The Paper, unauthorized reproduction prohibited +1 Save I want to report #Google #Gemini View more