← Back

Google drops Gemini 4 Neon to steal the AI coding crown

Original version ·

Big tech is done pretending that conversational chatbots will build entire startups from a single prompt. Enter the middle child built specifically to juggle terminals, APIs, and actual software without losing its digital mind halfway through.

Sitting neatly between the heavy-duty Carbon and the lightweight Hydrogen, Gemini 4 Neon aims squarely at autonomous agent tasks. Rather than drafting poetry or polite emails, it manipulates developer tools, chains multiple actions together, and swallows up to one million tokens of code context without blinking.

Users can dial its cognitive horsepower up or down via four distinct reasoning gears, running from low all the way to xhigh depending on how much deep contemplation a bug actually deserves.

On the competitive scoreboard, Neon snatched first place on the Artificial Analysis Coding Agent Index with 58.6 points, alongside taking third place on the general Intelligence Index at 61 points. It chewed through Terminal-Bench 4.1 at 65.2% and FrontierCode 2.0 at 46.2%.

Beyond raw bash commands, the model flexes its problem-solving on corporate workflows and compliance tasks, notching 56.3% on the Zapier AutomationBench.

For cyber defense, a specialized security variant scored 90.5% in SEC-bench Pro and 76% on AutoBaxBench, hunting vulnerabilities and patching exploits exclusively for vetted specialists inside the Fairwind program.

Access is live across the Gemini API and Google AI Studio, priced at $3 per million input tokens, $15 per million output tokens, and a tiny $0.15 for cached inputs.

Software engineers spent years joking that an artificial mind could never replace them because clients never know what they actually want. Now autonomous agents are learning to execute workflows, patch zero-days, and refactor entire repos faster than junior developers can even read the onboarding documentation.

Source: Google

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

0/24
  1. No comments yet.