What is new
Carbon builds on Gemini 4 Argon, which launched two weeks earlier. DeepMind describes Carbon as its strongest model, with what it calls state-of-the-art cybersecurity and long horizon capabilities. Those are the company's own words about its product.
On WebDev Arena, Carbon lands at position #4 with 1768 points, a gain of 89 elo points. DeepMind also says the Chrome team reported that Carbon performed more than 10X better than another lab's large frontier model when generating high quality patches.
Safety and benchmark claims
On Terminal-Bench 4.0, DeepMind says successful cheating fell to 0.0% and cheating attempts dropped to 0.4%. The company reports that high-severity misalignment alerts in live coding environments with internal teams dropped by almost 80%.
On Gray Swan's Indirect Prompt Injection Benchmark, Carbon reached a 0.0% attack success rate, against 0.7% for Argon. DeepMind notes that external benchmarks were evaluated without guardrails. The company has not said how the results were verified independently.
Access and pricing
Starting the week of the announcement, Carbon goes to a subset of cyber defenders, with phased access for more trusted partners through the Fairwind Program. Wide availability to developers, businesses and consumers is planned later through product surfaces, but no date was given.
The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input token price. That is the same introductory price as Argon. DeepMind has not said how long the introductory rate lasts or what the next price will be.
DeepMind says it is taking part in the U.S. government's voluntary process for pre-release model testing. It has not said whether the model or its weights will be released openly.
What it means for creators and sellers
Most video creators, social media managers and sellers cannot use Carbon yet: access starts with cyber defenders and broader rollout is undated. The published token prices may be worth noting for later budgeting, but the rate is described as introductory.
The launch material centers on security, patching and coding work rather than video or image generation. It is worth checking the official documentation once wider access opens to see what applies to your workflow.
How this was made: the text was written by AI from the source material, checked against the source and reviewed by an editor. The illustration was generated by AI for AISTETIKA and is not a frame from the story.
