AI

Google introduced Gemini 4 Neon, a model for coding and analysis that also comes to AI Pro subscribers

The fourth model in the Gemini 4 series sits between Carbon, built for the hardest tasks, and the faster Hydrogen

Image: Google

Google has introduced Gemini 4 Neon, the fourth model in the Gemini 4 series in less than two months. According to the company, Neon is designed for complex coding, business analysis and science, and it sits between Carbon, which is meant for the hardest and longest tasks, and Hydrogen, which is built for fast everyday work.

Developers can use it through the API in Google AI Studio, but only on the paid tier, while businesses get it through Gemini for business. Regular users can pick it in the Gemini app and in Antigravity if they pay for Google AI Pro or Ultra. That means that, unlike Carbon, which is Ultra-only, Neon also comes to the cheaper plan.

API pricing is $3 per million input tokens and $15 per million output tokens, with a one-million-token context window. That’s 25% less than Carbon’s regular price, although Carbon is cheaper until November 30 thanks to its introductory pricing. According to Artificial Analysis, Neon at the xhigh level uses about 15% more output tokens than Carbon at high, but it still costs about 17% less per task than Carbon at its regular price.

On the Artificial Analysis leaderboard, Neon at xhigh scores 61 points and ranks third, behind Claude Fable 5.2 and GPT-6.2 Astra, and one point ahead of Carbon. On the Vals Index, it’s second with 70.9%, again behind Fable 5.2.

In Google’s own table, Neon is compared with Claude Opus 5.5 and GPT-6.2 Sol and comes out on top in most tests. Still, Opus 5.5 beats it on Terminal-Bench 4.1 (65.8% vs. 65.2%) and on the FrontierCode 2.0 coding test (47.3% vs. 46.2%), while GPT-6.2 Sol leads on Chartography, which measures chart reading (75.1% vs. 73.5%). Fable 5.2 and Astra aren’t in the table.

Google also announced a separate Neon for cyber version, which vetted cyber defense experts will get in the coming days through the Fairwind program, the same program through which Carbon itself was first made available. According to the company, Neon for cyber writes more secure code and fixes vulnerabilities better than Carbon, while on the other cybersecurity tests it delivers results on par with top models using fewer tokens and at a lower cost.

It looks like Neon offers Carbon-level results at a lower price, and AI Pro subscribers get a model that, at least on Artificial Analysis, is well ahead of Argon.