Google’s new frontier model, Gemini 4 Carbon, brings some welcome improvements and while it doesn’t top the charts in a new benchmark, it does provide much more bang for your buck.
SemiAnalysis, an AI research and analysis firm, recently published its new frontier-level benchmark, GDPx which includes numbers for Google’s Gemini 4 Carbon model. The leaderboard includes every major frontier model from the likes of OpenAI, Anthropic, and Meta and details both the score and cost for each model.
The benchmark includes 206 tasks across finance, materials science, cybersecurity, software engineering, legal, biology, and several other domains (topics).
Topping the results is Anthropic’s Claude Opus 5.5, but Google’s Gemini 4 Carbon is very close behind. The difference between the two is relatively minimal, but the cost is the biggest factor to consider here. Where Claude’s cost came out to around $51 per task, Gemini was just over $16 per task. Gemini also beat the next model on the list, GPT-6 Astra, which scored lower while costing more than both at about $54 per task.
The report notes that taking out cybersecurity, biology, and alignment domains shuffles the results. Safety guardrails in Claude and GPT models puts GPT-6 Astra at the top of the list with GPT 6.1 Sol and Claude Sonnet 5.5 outranking Gemini 4 Carbon. SemiAnalysis says that this better aligns with their “subjective impressions” of each model.
As for Google’s existing frontier model, Gemini 4 Argon, it ranks in eighth place on both lists, beating Meta’s Muse Wonder, but below Wonder Full. Argon’s costs in this particular benchmark are only slightly lower than Carbon.
Both of Google’s models score better than OpenAI and Anthropic models when focused on cybersecurity, but do fall behind Meta’s Muse models, which the benchmark found to be the best “by a meaningful margin.”
Google announced Gemini 4 Carbon on October 28 and touted “significant gains,” but the public release date has not been confirmed just yet. It is “coming soon” for most customers. Google’s own benchmark, below, beat GPT-6 Astra and Opus 5.5 in many aspects.
More on Gemini:
Follow Ben: Twitter/X, Threads, Bluesky, and Instagram
FTC: We use income earning auto affiliate links. More.