Gemini 4 Carbon ranks second on SemiAnalysis GDPx at just over $16 per task, behind Claude Opus 5.5

Google's Gemini 4 Carbon finished just behind Anthropic's Claude Opus 5.5 on GDPx, a new frontier benchmark from SemiAnalysis, at just over $16 per task. Claude Opus 5.5 topped the leaderboard at around $51 per task, according to a 9to5Google report published November 2, 2026.

What GDPx measures

SemiAnalysis, an AI research and analysis firm, built GDPx with 206 tasks across finance, materials science, cybersecurity, software engineering, legal, biology and other domains. The leaderboard covers every major frontier model from companies including OpenAI, Anthropic and Meta, and lists both a score and a cost for each model.

Results

• Claude Opus 5.5 (Anthropic): first place, around $51 per task.
• Gemini 4 Carbon (Google): a close second, just over $16 per task. 9to5Google describes the score gap to Opus 5.5 as relatively minimal.
• GPT-6 Astra (OpenAI): scored lower than Gemini 4 Carbon at about $54 per task, the highest cost of the three.
• Gemini 4 Argon (Google): eighth on both lists. It beats Meta's Muse Wonder but ranks below Wonder Full. Its cost in this benchmark is only slightly lower than Carbon's.

Rankings shift without three domains

The SemiAnalysis report notes that removing the cybersecurity, biology and alignment domains reshuffles the results. The report attributes this to safety guardrails in Claude and GPT models. With those domains excluded, GPT-6 Astra ranks first, and GPT 6.1 Sol and Claude Sonnet 5.5 both rank above Gemini 4 Carbon. SemiAnalysis says this ordering better matches its "subjective impressions" of each model.

On cybersecurity, both Google models score better than OpenAI and Anthropic models. Meta's Muse models score higher still: the benchmark found them to be the best "by a meaningful margin."

Release status

Google announced Gemini 4 Carbon on October 28 and touted "significant gains." The public release date has not been confirmed. The model is "coming soon" for most customers. Google's own benchmark results, according to the company, show Gemini 4 Carbon beating GPT-6 Astra and Opus 5.5 in many aspects.

What this means

On GDPx, cost per task separates Gemini 4 Carbon from the two models above it. Carbon sits just behind Claude Opus 5.5 at just over $16 per task, while Opus 5.5 costs around $51 and GPT-6 Astra about $54. The domain-exclusion result shows how sensitive the leaderboard is to what is counted: with cybersecurity, biology and alignment removed, three other models rank above Carbon. Teams comparing models should check which domains match their workloads. Because the public release date for Gemini 4 Carbon is unconfirmed, these figures describe benchmark runs, not generally available access.