⚡ Uncle Cat AI Radar
ModelsAgentsIndustry

Google launches Gemini 4 Neon for faster agentic work

Google has released Gemini 4 Neon across its consumer, developer and coding surfaces, emphasizing faster completion of complex multi-step tasks.

What changed

Google launched Gemini 4 Neon on November 18, making it available to Google AI Pro and Ultra subscribers, paid Gemini API users, Google AI Studio, Antigravity and GitHub Copilot. The model sits between Gemini 4 Carbon and Hydrogen in Google’s lineup and is positioned for software engineering, enterprise knowledge work and scientific tasks.

Google describes Neon as a frontier model optimized for end-to-end task completion rather than isolated answer quality. In Gemini Tasks, it can handle longer, multi-step jobs; in Copilot and Cursor, it is aimed at agentic coding and tool use. Artificial Analysis reported that Neon ranked third on its Intelligence Index at the xhigh thinking setting, while Arena data placed Google models in the top three positions in text evaluation.

The commercial proposition is speed and economics. Artificial Analysis estimated that Neon completed an Intelligence Index task in roughly 21 minutes, about 40% faster than Gemini 4 Carbon, while costing about $6.44 per task at the tested setting. Cursor reported a 63.7% score on CursorBench 4.1, compared with 57.3% for the prior Gemini model.

Why it matters

Neon is less a single-model announcement than a distribution push: Google is putting one agent-oriented model simultaneously into consumer chat, coding products, cloud APIs and third-party developer tools. That makes performance-per-minute and tool reliability as important as benchmark rank. The unresolved question is whether Neon’s speed advantage survives real production workloads, where long contexts, permissions and failed tool calls often dominate total cost.

Uncle Cat take: Neon’s 21-minute task time matters more than its third-place ranking; Google is pricing agentic throughput as a product feature.

Sources