Gemini 4 Carbon tops Claude Opus 5.5 in benchmarks, hallucinates less

Google released Gemini 4 Carbon last week. According to benchmarks, the new model topped AI intelligence rankings, narrowly beating Anthropic’s Claude Opus 5.5.

The same tests show Gemini 4 Carbon is the least likely of the frontier models to hallucinate. That is a benchmark result, not an independently confirmed fact, and positions in such rankings depend on which tests are used.

Hallucinations are confident answers built on invented facts. They are one of the main complaints about all language models, so a claim of the lowest hallucination rate matters to Google as much as first place for intelligence.

KRAKOBIT already covered the release on October 29: Google’s Gemini 4 Carbon Leads Vulnerability Benchmarks. A comparison of ChatGPT, Claude and Gemini subscriptions against local models is in the article AI subscriptions in 2026.

Source: Tom's Hardware