# Gemini 4 Carbon Beats Claude Opus 5.5 On Intelligence While Being 23% Cheaper > Source: > Published: 2026-10-28 17:36:31+00:00 Google is back with a bang in the AI race with Gemini 4 Carbon. Google’s frontier model lineup has reached a new milestone as independent evaluator Artificial Analysis crowned Gemini 4 Carbon as the new #1 model on its Intelligence Index v5. Reaching an overall intelligence score of 59, Google’s latest entry managed to edge out major market rivals, including Anthropic’s flagship Claude Opus 5.5. Crucially for enterprise buyers, Gemini 4 Carbon achieved this top rank while delivering significantly better unit economics, costing $7.76 per Intelligence Index task compared to $10.04 for Claude Opus 5.5—a 23% savings for comparable overall intelligence. The model’s efficiency advantage stems largely from its streamlined generation process and token economy. Gemini 4 Carbon consumes roughly 123k output tokens per evaluated task, which is 29% fewer tokens than the 173k required by Claude Opus 5.5. Standard pricing for Carbon sits at $4 per million input tokens and $20 per million output tokens, which matches Gemini 4 Argon while scoring 9 points higher on the main Intelligence Index. Google is also offering a temporary 50% launch discount, accompanied by a steep 95% discount ($0.20 per million tokens) on cached input processing to keep enterprise deployment overhead predictable. In specific benchmark evaluations, Gemini 4 Carbon leads the Intelligence Index v5 across four of its nine core categories, driven by standout performances in HLE-Diamond, Terminal-Bench-Science, AutomationBench-AA, and AA-Omniscience v1.1. On the updated AA-Omniscience test, Carbon set a new high-water mark with a score of 54—six points clear of Claude Opus 5.5. The model achieved an accuracy rate of 58% alongside an 11% hallucination rate, which remains among the lowest in the market compared to the 50% hallucination rate observed in both Claude Opus 5.5 and OpenAI’s GPT-6 Astra. Beyond pure logic and reasoning, Google made notable technical strides in agentic coding performance, an area where the company has historically trailed Anthropic. Paired with the Antigravity CLI, Gemini 4 Carbon reached a score of 59 on the Artificial Analysis Coding Agent Index, finishing just three points behind Claude Code paired with Claude Opus 5.5. On broader agentic knowledge work evaluated via AA-Briefcase, Carbon posted a 126-point improvement over Gemini 4 Argon to reach 1609 Elo, fueled by major leaps in presentation and analytical quality. While still trailing specialized productivity models like Grok 4.7 and Kimi K3.1 Pro on knowledge workflows, Gemini 4 Carbon’s blend of high reasoning accuracy and lower execution cost marks a strong competitive shift for Google DeepMind.