Google releases Gemini 4 Carbon — new model leads in three agentic...

Google releases Gemini 4 Carbon — new model leads in three agentic benchmarks

Google released Gemini 4 Carbon on November 10, 2026 — a flagship for complex multi-step programming, agentic tasks, and scientific reasoning, which, according to the company, leads in CursorBench 4.0, Perplexity WANDR, and Zapier AutomationBench at a task cost more than twice lower than the nearest external model.

Compiled by Sergey KostenchukPublished 2026-11-10Updated 2026-11-10

2026-11-10 Research Anthropic

🤖 Google releases Gemini 4 Carbon — leader in three agentic benchmarks

The Gemini 4 flagship for complex multi-step code, agentic tasks, and scientific reasoning. According to Google, in AutomationBench, Carbon scored 53.9% compared to 51.3% for Argon, 44.7% for Claude Sonnet 5.5, and 41.4% for GPT-6 Astra.

🌍 The most expensive segment — agentic pipelines and debugging — is being attacked by Google with the argument of "twice as cheap as the nearest external model," pressuring Claude and GPT-6. In Antigravity, a preview of /swarm is open: one prompt launches a group of agents that write and test code in parallel.

👤 Until November 30, the rate is $2/$10 per 1 million tokens, then $4/$20 — a window for testing agentic pipelines in AI Studio and Antigravity. Researchers can apply for 6 months of free access to Argon and Carbon.

Source 1: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/

Source 2: https://zapier.com/benchmarks/gemini-carbon-4-report

Sources