Gemini 4 Carbon Beats Claude and GPT-6: Why Is Google Holding It Back?

Google's Gemini 4 Carbon has taken first place in the Artificial Analysis Intelligence Index, scoring 59 to edge out Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra. The achievement sent Alphabet shares up approximately 1.5% on Wednesday. Despite its top ranking, most paying customers cannot yet access this advanced AI model.

Restricted Access

Google is implementing a phased release strategy for Gemini 4 Carbon, prioritizing safety. The model is currently available only to vetted cyber defenders through its Fairwind Program. Developers, businesses, and consumers will follow, though Google has not specified when wider access will begin.

The company's benchmark tests reveal Carbon's significant advantage in security-related tasks. On ExploitGym, which evaluates a model's ability to identify usable software vulnerabilities, Carbon scored 54.8%, substantially higher than GPT-6 Astra's 42.4% and Claude Opus 5.5's 34.5%. Google conducted these tests with safety filters disabled.

Pricing Structure

Google has announced pricing for Gemini 4 Carbon, charging $2 per million input tokens and $10 per million output tokens initially. These prices will later double to $4 and $20 respectively. This contrasts with OpenAI's recent price cuts for GPT-6.1 Terra to $1 and $5 per million tokens.

While Carbon leads in security benchmarks, rivals still outperform in certain areas. On FrontierSWE 2, a coding assessment, GPT-6 Astra scored 65.5% and Claude Opus 5.5 scored 62.3%, compared to Carbon's 61.2%. Google has not provided a timeline for when the higher pricing will take effect or when the model will be available to the general public.