Google's Gemini 4 Carbon looks like a diamond in the rough

If you pay for AI models, Google's new offering matches a competitor's price for its most powerful tier.

Google has launched Gemini 4 Carbon, a more powerful version of its AI model, just two weeks after releasing Gemini 4 Argon. Initially, Carbon will be offered at the same price as Argon: $2 per million input tokens and $10 per million output tokens. However, this introductory pricing is temporary, with plans to increase to $4 per million input tokens and $20 per million output tokens once the promotional period ends. This new model directly competes with Anthropic's Claude Opus 5.5, which is priced identically at $4 and $20 per million tokens. Gemini 4 Carbon has demonstrated strong performance in cybersecurity benchmarks, including SEC-bench Pro, ExploitGym, and CWE-bench, and shows significant improvement in generating secure code and code patches compared to its predecessor. Google CEO Sundar Pichai stated that their first-party model APIs are processing nearly 32 billion tokens per minute. Google plans to make Gemini 4 Carbon broadly available as soon as possible, with an initial release this week to select cybersecurity defenders.

Google's aggressive pricing strategy for Gemini 4 Carbon, matching Anthropic's Opus 5.5, signals an intensifying battle for market share in the high-end AI model space. By offering superior performance in critical areas like cybersecurity at a competitive price point, Google aims to disrupt Anthropic's momentum ahead of its IPO. This move also highlights Google's commitment to leveraging its substantial infrastructure and AI expertise to challenge established leaders, potentially forcing other players to reconsider their own pricing and performance benchmarks.

The rapid iteration and release of Gemini 4 Carbon, following closely on the heels of Argon, underscores the fierce pace of development in the AI sector. Companies are now expected to deliver continuous improvements in model capabilities, particularly in specialized domains like cybersecurity. The broader availability of Carbon, starting with cybersecurity defenders, suggests a strategic focus on enterprise solutions and a clear roadmap for scaling access, indicating a future where advanced AI tools become increasingly integrated into specialized security workflows.