Google Unveils Gemini 4 Carbon and Claims the Intelligence Index Crown
Google has officially released its latest artificial intelligence model, Gemini 4 Carbon, capturing the top spot on the Artificial Analysis Intelligence Index V5 with a score of 59, telset.id reported. The release marks a major comeback for the tech giant following an irregular rollout schedule throughout the year. The high-tier model edges out Anthropic’s Claude Opus 5.5, which secured a score of 58.
Despite claiming the intelligence crown, the new model remains locked away from most paying customers. Google announced that access is currently restricted to verified cyber defenders and select partners through the Fairwind Program. Alphabet shares rose about 1.5% on the day of the announcement.
Weighing Performance Against Competitors
Gemini 4 Carbon achieved a score of 54 on the AA-Omniscience v1.1 benchmark, which measures information retention and hallucination rates. According to telset.id, this beats Claude Opus 5.5’s score of 48. However, Google’s internal data shows that while Carbon answers more questions correctly than its predecessor Gemini 4 Argon—reaching 58% accuracy compared to Argon’s 48%—it also exhibits a higher hallucination rate of 11%, up from Argon’s 4%. Pintu reported that the model also scored 88.2% on SEC-bench Pro for identifying software vulnerabilities.
Full benchmark results for Gemini 4 Carbon with high reasoning pic.twitter.com/p4aPtYADZd
Pricing remains a competitive battleground for frontier AI models. Telset.id reported that Gemini 4 Carbon costs US$7.76 per task, which is cheaper than Claude Opus 5.5. At launch, Google priced the model at US$2 per million input tokens and US$10 per million output tokens, with a 95% discount for cached input tokens, CNBC Indonesia reported. OpenAI lowered the price of its competing GPT-6.1 Terra model to US$1 and US$5 on October 26.
Cybersecurity Focus and Phased Rollout Strategy
Koray Kavukcuoglu stated that the new model brings significant enhancements to cybersecurity, agentic tasks, and software engineering, CNBC Indonesia reported. Google’s internal testing demonstrated that Carbon scored 54.8% on ExploitGym and 71% on CWE-bench for fixing code vulnerabilities. The tech giant’s Chrome team reported that the model’s performance in generating high-quality software patches is more than 10 times that of frontier AI models from other labs.
Google chose a deliberate rollout approach similar to Anthropic’s 2025 launch of Mythos, collaborating with the United States government for pre-release safety evaluations. Google ran external cybersecurity evaluations without safety filters. The company has not yet announced a timeline for when general developers, businesses, and consumers will gain broad access to the model, according to Pintu.
Details on release date, access, and pricing
When was Gemini 4 Carbon announced?
Google DeepMind announced Gemini 4 Carbon on October 28, 2026, roughly two weeks after the release of Gemini 4 Argon, Pintu reported.
:quality(82)/8e3372e3ac9a451bb7af442192ac5e22 "Photo: Beincrypto")
Who currently has access to Gemini 4 Carbon?
Access is currently limited to verified cyber defenders and select partners participating in Google’s Fairwind Program, according to telset.id.
How much does Gemini 4 Carbon cost to use?
Google introduced the model at a promotional rate of US$2 per million input tokens and US$10 per million output tokens, offering a 95% discount for cached input tokens, CNBC Indonesia noted.
More on Artificial Analysis Intelligence Index
Google’s Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5, and second only to Claude Opus 5.5 on the Artificial Analysis Coding Agent Index Gemini 4 Carbon scores 59 on Intelligence Index v5, driven by leading performance in complex reasoning,… pic.twitter.com/gSH65FSKzL
Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned. pic.twitter.com/Ikk9YedjCQ
Related reading