Google’s Gemini 4 Carbon Claims the AI Crown—But You Still Can’t Use It
Google’s Gemini 4 Carbon Claims the AI Crown—But You Still Can’t Use It
Two weeks after introducing Gemini 4 Argon, Google announced Gemini 4 Carbon, its latest frontier model, framing the rapid follow-up as an acceleration in the race against OpenAI and Anthropic. The company says Carbon raises the bar in coding, agentic work and scientific analysis—especially cybersecurity—but is withholding broad access while it tests the risks.
Sundar Pichai cast the launch as evidence of momentum, saying Carbon “shows the pace at which we’re pushing the frontier” and ranks first on Artificial Analysis’s updated Intelligence Index. Demis Hassabis struck the same note, calling the model state-of-the-art on key benchmarks and emphasizing its combination of “frontier capabilities” with safety and security research.
Google’s headline figures are muscular. It says Carbon is more than twice as likely as Argon to produce code without vulnerabilities, while Chrome engineers found it generated high-quality patches more than 10 times better than an unnamed competing frontier model. In its Grey Swan assessment, Google also says Carbon resisted every prompt-injection attempt. But the company’s own deliberative rollout is an acknowledgment that stronger cyber and autonomous-task capabilities demand restraint.
The benchmark story is less absolute than the victory lap. Carbon scored 59 on Artificial Analysis’s Intelligence Index v5, ahead of Claude Opus 5.5 and GPT-6 Astra, according to coverage of the results. Yet it does not lead every test: on the AA-Briefcase knowledge-work benchmark, it trails models from SpaceXAI, Meta and Moonshot. The reported gaps among leading systems are often only a few points.
For now, Carbon is available only to selected users, with wider partner access planned through Google’s Fairwinds cyber-defense program before a broader release. Developers and AI Ultra subscribers are expected to be first in line. The split-screen moment is clear: Google is declaring a comeback, while asking the public to take its strongest claims on trust. Even Elon Musk’s response to Pichai’s announcement was simply: “Impressive.”
Write a comment