Key Findings
Google’s Gemini 4 Carbon is positioned as the company’s most high-performance model to date, with a primary focus on enhancing cyber defense capabilities and the processing of complex tasks over extended periods. In the Artificial Analysis Intelligence Index (v5.0), Carbon scored 59 points, taking the top position and marking a 9-point increase over the 50 points scored by Gemini 4 Argon.
Technical Details
The model demonstrates significant improvements in autonomous software development and cybersecurity. According to internal evaluations, the success rate for generating secure code is more than double that of Argon. Carbon is capable of operating autonomously for several days to investigate and fix bugs within large-scale codebases.
Safety and reliability metrics show the following improvements over the Argon model:
- Hallucinations were reduced by approximately half.
- Critical alignment issues in coding environments decreased by approximately 80%.
- The attack success rate in the “Gray Swan IPI,” which measures resistance to indirect prompt injection, dropped from 0.7% to 0.0%.
Background and Context
Gemini 4 Carbon follows the release of Gemini 4 Argon, which was announced on September 30 and became generally available on October 13. Logan Kilpatrick of Google DeepMind noted that Carbon represents a significant advancement over Argon in terms of coding usability, alignment, and long-term task processing.
Pricing for the API remains consistent with Argon. The introductory price is set at 2 dollars per 1 million tokens for input and 10 dollars for output; after the introductory period, these rates increase to 4 dollars for input and 20 dollars for output.
Significance and Outlook
Google is implementing a phased rollout, starting with select cyber defense organizations participating in the “Fairwind Program.” General availability for developers, enterprises, and individual users will follow based on safety verification and feedback. To mitigate national security risks and potential misuse for cyberattacks, Google is collaborating with the AI Safety Institute (CAISI) under the U.S. National Institute of Standards and Technology (NIST) for pre-release safety evaluations.
Source: https://www.itmedia.co.jp/news/article/2610/29/2000002690/
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
