Google places Carbon at the top of its advanced AI models
In brief
- Google replaces Argon with Carbon as its new flagship AI model, leading 15 of 17 public benchmarks.
- Carbon cuts severe misalignment alerts by 80 % and achieves a 0 % cheating rate on Terminal-Bench 4.0.
- Public release expected in two weeks, but Google provides no details on pricing or precise timing.
Google is replacing its recently unveiled Argon model, launched at the end of September, with Carbon, presented as its new flagship model. According to the company, Carbon now leads 15 out of 17 public benchmarks, outperforming Argon and other competitors like GPT-6 Astra or Opus 5.5.
Google’s figures highlight significant improvements, including an 80 % reduction in severe misalignment alerts and a 50 % drop in unsupported responses compared to Argon. Carbon also achieves a 0 % cheating rate on Terminal-Bench 4.0. However, Google has not disclosed its public release date or final pricing, though the model currently follows Argon’s pricing structure.
Carbon is currently limited to a small group of cybersecurity specialists via the Fairwind program, with gradual access expansion planned. Its public launch could occur within two weeks, but Google remains vague on details. The Gemini lineup lacks clarity, as Argon remains in the catalog despite Carbon’s promotion, raising questions about the future of both models.
What it changes for you
- CybersecurityCybersecurity professionals will be able to test Carbon through the Fairwind program, with broader access planned, though internal tests lack safeguards.
This summary gives you the essentials. The full article, its figures and quotes are on Next.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error