AI

Google DeepMind Releases Gemini 4 Neon at $3 per Million Input Tokens

502 words, 3 min

Source: deepmind.google

google deepmind releases gemini 4 neon at 3 per million inpu openverse 30243753 ac7b 4e64 a463 f2de93a794e2

Image: Public domain via Openverse

Google DeepMind released Gemini 4 Neon on November 18, 2026, a model that sits between Gemini 4 Carbon and Gemini 4 Hydrogen in its lineup. Google DeepMind said in the company's announcement that Neon is built for complex software engineering, knowledge work and science, with an emphasis on finishing tasks quickly.

What Google announced

Neon is the middle tier of the Gemini 4 family. Google describes it as frontier-level intelligence at a lower cost than Carbon, which makes it the model the company is steering developers and businesses toward for high-volume work. The rollout began on the day of the announcement.

Launch coverage lists a context window of roughly one million tokens. Default output in standard synchronous responses is reported at about 64,000 tokens, and a background long-decode mode can reach further. Those figures vary slightly between outlets, so check Google's documentation before building around them.

Advertisement

Availability and pricing

Developers can reach Neon through the Gemini API and Google AI Studio. Google is also offering it in Gemini for business and in its Antigravity coding environment. In the Gemini app, access is limited to Google AI Pro and AI Ultra subscribers. Google has not said whether free Gemini users will get Neon.

Reported paid-tier pricing is $3 per million input tokens and $15 per million output tokens. Cached input is listed at $0.15 per million tokens, and batch processing is reported at half the standard rate. Neon is not offered on the API free tier.

A separate cyber-focused variant is going to vetted defenders through Google's Fairwind Program rather than to the general public.

Agent workloads cost more than the headline rate

Neon is the fourth Google model release in about a month. The company has been filling out the Gemini 4 family quickly, and Neon is the first tier priced for teams that run agents or coding assistants at scale. Output tokens cost five times as much as input tokens, so agent workloads that generate long responses will cost more than the headline $3 figure suggests.

Most of the benchmark results circulating at launch come from Google's own materials. Independent evaluators have started publishing scores, but they are early, and Carbon's own numbers differ between sources, which makes cross-model comparisons unreliable for now.

Independent tests and billing will settle the Neon case

Three things will show whether Neon holds up outside Google's demos. The first is how third-party evaluations of coding and agent tasks compare with Google's figures once more of them are published. The second is whether Google extends Neon to free Gemini app users. The third is whether the batch discount and cached-input rate apply the same way in production billing as they do on the pricing page.

Teams already on Gemini 3.1 Pro or the earlier Argon model should run their own test sets against Neon before switching, since cost per completed task depends on how many tokens a model uses to reach an answer, not only on the per-token rate.

Featured image: Public domain via Openverse

Advertisement