Gemini 4 Carbon Comes Close to Claude Opus 5.5 at a Third of the Cost in SemiAnalysis Test

SemiAnalysis's GDPx benchmark tests models on 206 tasks across coding, finance, legal work and biology. According to Android Headlines' account of the results, Claude Opus 5.5 ranked first overall at about $51 per task. Gemini 4 Carbon came close behind at just over $16 per task. OpenAI's GPT-6 Astra scored lower than both and cost about $54 per task. Google's Gemini 4 Argon, released about two weeks before Carbon, placed eighth and is priced slightly below it.

For companies buying AI services, cost per task is a more practical measure than cost per token, because it captures how many tokens a model consumes to finish a job. On this test, Google is offering close to top-ranked performance at a much lower price.

The ranking changes with the task mix. With cybersecurity, biology and safety-guardrail tests removed, GPT-6 Astra moves to the top, and OpenAI's GPT-6.1 Sol and Anthropic's Claude Sonnet 5.5 both pass Gemini 4 Carbon. SemiAnalysis said that view is closer to how engineers experience the models in everyday coding. In specialised security tasks, both Gemini 4 models beat the OpenAI and Anthropic entries, and only Meta's Muse models scored higher.

The price gap may not last. Google launched Carbon at an introductory rate of $2 per million input tokens and $10 per million output tokens, the same as Argon, according to a report on the launch. After an unspecified period, the price is due to rise to $4 and $20, which matches what Anthropic charges for Claude Opus 5.5. The model was first released to selected cybersecurity defenders, and access is set to widen through Google's Fairwind programme.

Google has not given a date for the end of the introductory pricing, and that date will decide how long Carbon's cost advantage on this benchmark holds.