Signal

Artificial Analysis: Gemini 4 Neon trails only Claude Fable 5.2 and GPT-6.2 Astra on the Intelligence Index and rivals Opus 5.5 on coding at a much lower cost

First reported by Artificialanalysis ·

The signal ●●●○ Compiled by AI from Artificialanalysis and Techmeme
Why you might care

Coding assistant costs fall by nearly half, and performance now matches premium models.

What happened

Google DeepMind has released Gemini 4 Neon, its fourth new model in seven weeks, achieving a score of 61 on the Artificial Analysis Intelligence Index. This places Neon just behind GPT-6.2 Astra and Claude Fable 5.2, but ahead of Gemini 4 Carbon. Neon offers a slight intelligence gain over Carbon at approximately 17% lower cost per task and a 40% reduction in time per task. The model also demonstrates competitive performance in coding agents, matching Claude Opus 5.5 and Claude Fable 5.2 on the Artificial Analysis Coding Agent Index, but at a significantly lower cost. Gemini 4 Neon features a 1M token context window, supports multimodal inputs (text, image, video, speech), and is available via Google's API and Gemini Enterprise Agent Platform, with Pro and Ultra subscribers accessing it in Antigravity and the Gemini app. Pricing for Neon is $3/$15 per 1M input/output tokens.

What it means

Gemini 4 Neon's lower pricing and faster output speed, despite using more tokens, highlight a strategic shift towards cost efficiency in model development. The 25% reduction in token prices and substantial speed improvements suggest Google is optimizing inference costs to make advanced AI more accessible. This move pressures competitors to reconsider their own pricing structures, especially in specialized areas like coding agents where Neon now rivals higher-priced options.

The model's advancements in visual reasoning and presentation quality, alongside its 1M token context window, indicate a push towards more capable and versatile multimodal agents. Its performance on benchmarks like the Multimodal Knowledge Agent evaluation, where it matches leading models, suggests a broadening of AI application beyond text-based tasks. This will likely accelerate the development of sophisticated AI tools that can process and generate insights from diverse data types.

AI-written summary. May contain errors.

AI Chips