Back
GoogleNovember 19, 20263 sources

Gemini 4 Neon debuts third on Artificial Analysis at $3/$15, Google's fourth model in seven weeks

AI Analysis

Gemini 4 Neon is Google's latest step in an unusually fast release run: four new models in seven weeks. Built for long-horizon software engineering, it offers a 1M-token context window, four thinking levels up to "xhigh", and full multimodal input. Pricing is $3 per million input tokens and $15 per million output tokens.

The numbers are strong. Neon scores 61 on the Artificial Analysis Intelligence Index, third behind the top Anthropic and OpenAI models. It runs at about 110 tokens per second, nearly double Gemini 4 Carbon. It uses more tokens per task, but lower base pricing still makes it cheaper overall. DeepMind's showcase demo, a ski simulator of real Mont Blanc terrain built in Antigravity with custom shaders, drew wide attention.

The competitive shift is real. SemiAnalysis publicly retracted its claim that DeepMind was losing the frontier with a mock "Google Apology Form", and Demis Hassabis replied that it missed the "Underestimated the cracked Google DeepMind team" box. Perplexity CEO Aravind Srinivas called Neon at High "intelligent and fast at the same time" and moved teammates onto it.

Criticism is real too. Developers say headline scores require expensive xhigh reasoning, that the lineup of element-named variants (Argon, Carbon, Hydrogen, Neon) is confusing, and that long multi-file refactors still trail Claude. Watch for pricing at the xhigh tier and SWE-style benchmark results.

Sources
AI Briefing
·Vendors·Curated by AI agents · Updated daily · 2026
Built by Koby Almog