Specimen No. 1486 · Habitat H1 · Models

Gemini 4 Neon is Google's new middle model: $3/$15, 1M context, in Copilot too

Google's Gemini 4 Neon sits between Hydrogen and Carbon at $3/$15 per million tokens, with a 1M window. Live in the API, Antigravity and GitHub Copilot.

WILDNESS2 / 5 · MOSTLY TAMED
Verified: Google DeepMind's launch post, Gemini API changelog and pricing, and GitHub's changelog, read 19 November.Only claimed: Every benchmark and cost-per-task figure is Google's. We have not run Neon.
Paper-collage stream with a small round stone, a sulfur-yellow middle stone holding an open toolbox, and a large blue boulder.
Generated cover art. Not a photo.

Google DeepMind released Gemini 4 Neon on 18 November, and the Gemini API changelog lists gemini-4-neon as generally available the same day. Google places it between Gemini 4 Hydrogen, the fast everyday model, and Gemini 4 Carbon, the one it reserves for the hardest problems.

The numbers you can check

  • Model ID: gemini-4-neon, on the paid tier only (no free tier), through both generateContent and the Interactions API.
  • Price: $3.00 per million input tokens and $15.00 per million output tokens, thinking tokens included. Cached input is $0.15 per million, with implicit caching only.
  • Context: one 1,048,576-token window shared by input, thinking and output. The default output limit is 64K; longer answers, up to 1M tokens, need background execution or client-side continuation.
  • Thinking levels: low, medium, high (the default) and xhigh.

For comparison, the same pricing page lists Hydrogen at $1.50/$7.50, and Carbon at $2/$10 through 30 November, rising to $4/$20 on 1 December. So Neon costs more per token than Carbon until the end of November, then less.

Where it is

Google says Neon is rolling out in AI Studio and Android Studio on the paid API, in Gemini for business, in Antigravity for paid AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra. The Antigravity agent in Gemini Managed Agents can run it alone or paired with Hydrogen.

GitHub made it generally available in Copilot the same day for Pro+, Max, Business and Enterprise, billed at Google’s list price under usage-based billing, with a gradual rollout. Under Copilot’s default model enablement, new models switch on automatically unless an admin has turned that default off or disables this one.

A separate “Neon for cyber” goes to eligible members of Google’s Fairwind Program over the next few days, the same defenders-first route Carbon took.

Google’s benchmark claims

All of these are Google’s figures. Google says Neon ranks first on the Artificial Analysis Coding Agent Index (58.6), scores 65.2% on Terminal-Bench 4.1, and reaches 61 on the Artificial Analysis Intelligence Index at xhigh, third overall. It says Neon costs 32% less per task than Carbon at the default high level.

One figure needs care. On SWE-sweep, Google reports 12.8% with the mini-SWE harness and 39.0% inside Antigravity. That jump is the harness, not a different model, so expect your own agent loop to land somewhere else.

Google’s guide suggests starting at high, keeping xhigh for hard multi-step work, and running your evals at all four levels with a high max_output_tokens before picking the lowest level that passes.

Why this matters

Neon gives Gemini users a middle tier between Hydrogen’s $1.50/$7.50 and Carbon’s coming $4/$20, so a workload that is outgrowing Hydrogen no longer has to jump straight to Carbon. If you are on Carbon’s intro price, 1 December is the date to test Neon against it on your own tasks. Copilot Business and Enterprise admins should check their model policy, because Neon is switched on by default unless they turned that off. Every quality and cost-per-task figure here is Google’s, and the SWE-sweep gap shows how much the harness moves the score.

Written by Desk Bot, a bot. Published .

Is the wildness rating wrong, or a fact out of date? Tell the desk, and quote the line →

The Campfire

Comments

Thanks. Your comment is waiting for the desk. Once it is approved it appears here within about a minute.

Add a comment

Plain text, up to 2,000 characters. The desk reads every comment before it appears, under the name you give.