Google announced its new frontier model, Gemini 4 Carbon, on October 28, and 9to5Google reports that it lands closely behind Anthropic's top offering in a new industry benchmark while undercutting competitors on price.
Key takeaways
- SemiAnalysis published a new frontier-level benchmark called GDPx detailing scores and costs.
- Anthropic's Claude Opus 5.5 tops the chart, with Google's Gemini 4 Carbon just behind.
- Gemini 4 Carbon costs just over $16 per task compared to around $51 for Claude.
- Google's existing model, Gemini 4 Argon, lands in eighth place on the leaderboards.
How does the Gemini 4 Carbon benchmark compare?
SemiAnalysis recently released its GDPx benchmark, which covers 206 tasks spanning finance, materials science, cybersecurity, software engineering, legal, biology, and other domains. The leaderboard features major frontier models from OpenAI, Anthropic, and Meta. Anthropic's Claude Opus 5.5 takes the top spot, though Google's Gemini 4 Carbon sits very close behind with a minimal difference in score.
Cost serves as a major differentiator in these findings. According to 9to5Google, Claude costs roughly $51 per task, while Gemini 4 Carbon runs at just over $16 per task. Gemini also outscores GPT-6 Astra, which costs about $54 per task and scores lower on the list.
What are the results of the frontier model benchmark?
Removing specific domains like cybersecurity, biology, and alignment changes the standings. Safety guardrails affect Claude and GPT models, pushing GPT-6 Astra to the top of that adjusted list, followed by GPT 6.1 Sol and Claude Sonnet 5.5 ahead of Gemini 4 Carbon. SemiAnalysis states this adjustment better matches subjective impressions of the systems.
Meanwhile, the older Gemini 4 Argon lands in eighth place on both lists, outperforming Meta's Muse Wonder while sitting below Wonder Full. Both Google models outperform OpenAI and Anthropic options on cybersecurity tasks, though Meta's Muse models win that category by a meaningful margin.
What happens next for Google?
Google touted significant gains when announcing Gemini 4 Carbon, though a public release date remains unconfirmed. It is coming soon for most customers, according to company statements. Google's internal tests showed its model beating GPT-6 Astra and Opus 5.5 across numerous aspects, setting the stage for its upcoming wider rollout.
Frequently asked questions
Gemini 4 Carbon is a new frontier model announced by Google on October 28, bringing notable improvements and lower costs compared to rival systems.
SemiAnalysis, an AI research and analysis firm, published the new frontier-level GDPx benchmark.
Gemini 4 Carbon costs just over $16 per task, making it significantly cheaper than competing models from Anthropic and OpenAI.
Anthropic's Claude Opus 5.5 took the top spot in the benchmark, with Google's Gemini 4 Carbon placing closely behind.
Google announced Gemini 4 Carbon on October 28, stating that a public release is coming soon for most customers.
Based on reporting by 9to5Google. Read the original article
