Google on November 18, 2026 added Gemini 4 Neon to the Gemini API as a generally available model. The identifier is gemini-4-neon, and the entry appears in the API release notes and on the Gemini API models page. Google describes it as a model with frontier-level intelligence and fast end-to-end task completion across complex software engineering, knowledge work and science workflows.

The listed specifications are specific. Neon is offered on the paid tier at $3.00 per million input tokens and $15.00 per million output tokens. It supports four thinking levels: low, medium, a default of high, and xhigh. Google gives it a unified 1M-token context window and says it can decode up to 1M tokens in a single response, while the models page shows a maximum output of 64K or 1M tokens and accepts text, image, audio, video and PDF input. Google points developers to a "What's new in Gemini 4 Neon" guide for migration details.

Neon is the fourth model in the Gemini 4 family to reach general availability in about six weeks, and the release notes make its place in the lineup easy to read. Gemini 4 Hydrogen, released on November 12 for the free tier and the paid tier, costs $1.50 and $7.50 per million tokens. Gemini 4 Carbon, which Google calls its strongest reasoning model, became generally available on November 10 at introductory pricing of $2.00 and $10.00 through November 30. Gemini 4 Argon is the efficient frontier option. On those numbers Neon sits above Hydrogen on price and, until the end of November, above Carbon's introductory rate, so its cost case depends on how much work it finishes per task rather than on the per-token list price alone.

The same November 18 entry carries a second change for developers who work with documents. The Gemini API now accepts Microsoft Word, PowerPoint and Excel files and OpenDocument files directly. It also lets prompts and function responses slice content: page ranges from documents, worksheet tabs and cell ranges from spreadsheets, time intervals from video and audio, and content layers such as page images, extracted text, formulas or video transcripts. Google says this lets applications process and pay for only the portion of a file they need.

Google has not published benchmark results for Neon in the release notes, and the notes do not compare it with Carbon on accuracy or speed. The positioning as a faster model near frontier quality is therefore the company's own description. Teams considering a switch from Hydrogen or Carbon will need to measure latency, token use and answer quality on their own workloads.

For builders, the practical change is a new default candidate for agentic coding and long-document jobs, together with finer control over what a request ingests. Because Neon uses the same thinking-level scheme as Hydrogen, with the added xhigh setting, existing Gemini 4 integrations should need only a model string change, though Google has not made that claim explicitly.

References