Google releases Gemini 4 Neon: input costs $3 and output $15 per million tokens
Google says Neon delivers frontier-level performance faster and at lower cost than the company’s larger models. Neon has begun rolling out on the paid Gemini API, to AI Pro and Ultra subscribers in the Gemini app and Antigravity, and gradually in GitHub Copilot.
In the Gemini 4 family, Neon sits between Hydrogen and Carbon. Google offers Hydrogen for fast everyday use and Carbon for the most demanding long projects. Neon has a context window of 1 million tokens, and cached input costs $0.15 per million tokens. It offers four levels of reasoning effort.
Neon costs twice as much as Hydrogen, according to 9to5Google. Until 30 November, Argon and Carbon are on promotional pricing of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 afterwards. Neon will then be 25 per cent cheaper than them, the site said. 9to5Google also reported that Neon replaces Argon in the Gemini app’s model menu. Google’s blog does not mention this.
Google’s blog says Neon scored 61 on the Artificial Analysis Intelligence Index at its highest reasoning effort, ranking third behind larger, slower models.
In an image posted on X, Google DeepMind compares Neon with Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6.2 Sol. In the image, Neon leads in most benchmarks, but Opus 5.5 is ahead on Terminal-bench 4.1 and FrontierCode 2.0, and GPT-6.2 Sol on Chartography. The comparison is Google’s own, and DeepMind’s page describes the methods. GitHub’s changelog presents Neon’s quality claims as Google’s.
Google says the SWE-sweep results show how much the harness matters. A harness is the program that steers an agent: it gives the model its tools and runs its steps. In the image, Neon scores 12.8 per cent on the shared mini-SWE harness, close to GPT-6.2 Sol (11.6 per cent) and well ahead of Opus 5.5 (3.6 per cent). In the providers’ own harnesses, Neon scored 39.0 per cent in Google’s Antigravity, Opus 5.5 10.1 per cent and GPT-6.2 Sol 8.3 per cent.
Google did not evaluate Neon’s dangerous capabilities with the full test suite, its model card says, because the increase in Neon’s capabilities over Argon fell below the company’s threshold. Argon’s evaluations therefore apply to Neon: for example, the possibility that the model could assist in serious cyberattacks cannot be ruled out. Google will publish the full safety report in the coming days.
Neon is available in Google AI Studio on the Gemini API’s paid tier and in Gemini for business. In Google Antigravity it is available to paying AI Pro and Ultra subscribers, and in the Gemini app to AI Pro and Ultra subscribers. In GitHub Copilot, the model is billed by usage at the provider’s list prices and is rolling out gradually to Pro+, Max, Business and Enterprise users. In Business and Enterprise organisations the model is enabled by default unless an administrator has turned off default enablement of new models or disables Neon separately in Copilot’s model policy.
Read more: How an agent works · Models