Third on the Index, Priced Under the Flagship: Google Ships Gemini 4 Neon at $3 and $15

Catalina Clarke, Editor-in-Chief, The Library Media

The Library Media — Science, Technology & Health · AI News & Intelligence Desk

Third on the Index, Priced Under the Flagship: Google Ships Gemini 4 Neon at $3 and $15

By Catalina Clarke

Share

Published November 19, 2026 · 9:59 AM PT

Google introduced Gemini 4 Neon on Wednesday, November 18, a new model in its Gemini 4 family built, in Google’s words, for “frontier-level intelligence with fast end-to-end task completion” across complex software engineering, analytical reasoning and scientific work. Neon is the model’s positioning as much as its specification: it sits between Carbon, the flagship Google reserves for the hardest long-horizon projects, and Hydrogen, its speed model for everyday work, and it is priced under the flagship’s standard rate while outscoring it on the index Google leads with.

The price is $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million — a 95% discount, on Google’s framing — and a context window of 1 million tokens. Carbon’s standard price in Google’s own family chart is $4 per million input and $20 output, and Hydrogen’s is $1.50 and $7.50. One honest complication belongs in the spec sheet: Carbon is currently on promotion at $2 and $10 through November 30, according to the footnote on Google’s chart, so for the rest of this month the flagship is briefly the cheaper model per token. Google’s counter-argument is cost per finished task rather than cost per token: at its default high thinking level, Google says, Neon pairs its index score with a 32% lower cost per task than Carbon.

The headline measurement is the Artificial Analysis Intelligence Index. Google reports that Neon scores 61 at its xhigh thinking level, ranking third overall, “just behind larger, slower models.” In Google’s family chart, that 61 edges Carbon’s 60 and Hydrogen’s 54. Neon offers four thinking levels — low, medium, high and xhigh — which Google describes as a finer dial between intelligence and token spend. Beyond the headline index, Google’s launch post reports Neon second on the Vals Index at 70.9%, first on that index’s Legal Research and Tax Agent benches, first overall on the Artificial Analysis Coding Agent Index at 58.6, and first on Zapier’s AutomationBench at 56.3%. Every one of those figures is Google’s reporting of the launch benchmarks; independent tracking of the model was incomplete on launch day, and readers should hold the rankings as the maker’s claims until outside measurement catches up.

Availability is paid-tier from the start. Neon is rolling out in the Gemini API in Google AI Studio on the paid tier — Google’s chart marks Neon “paid” where Hydrogen is listed for all tiers, so there is no free-tier access at launch — and developers can also build with it in Android Studio. Enterprises get it through Gemini for business, by API or model dropdown. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity, and Pro and Ultra subscribers can select it from the model dropdown in the Gemini app, where Google says Tasks can now run on Neon for complex, multi-step work. Google lists no batch-processing price for Neon in the launch post.

What Google says the model is for is finished work, not just answers. Enterprise material, as the post puts it, “lives in spreadsheets, PDFs, charts, audio, and video,” and Neon is pitched as synthesizing across all of those inputs inside its single million-token window and returning completed artifacts — reports, spreadsheets, slides and interactive dashboards, including a fully interactive spreadsheet with models, logic and formatting. For code, Google describes a model engineered for long-running agent workflows: multi-file debugging and refactoring across real repositories, with “targeted, surgical edits,” sustained through tool-heavy tasks such as adding a feature or running a major migration. Google also claims its lowest fabrication rate yet, measured in internal testing in Antigravity, and stronger writing on its own agentic writing evaluations.

A separate variant, Gemini 4 Neon for cyber, is not a general release. Google says it is rolling out “to trusted defenders in the coming days” through the Fairwind Program, with eligible members receiving access over the next few days and access requests routed through Cloud account teams. The cyber variant is built for defensive work — finding vulnerabilities, validating risk through exploitation, and patching — and Google reports state-of-the-art results for it on SEC-bench Pro, at 90.5%, and on its own WebExploitBench, at 51.26%, each at roughly half the cost of competing large models. On safety more broadly, Google says Neon maintains Carbon-level protections against severe misuse and prompt injection, and that at its highest thinking level with real-time safeguards it recorded a 0.0% attack success rate on Gray Swan’s indirect prompt injection benchmark, a level the post says was matched only by other Gemini models.

The honest summary is the one Google’s own chart tells. Neon is not the cheapest Gemini — Hydrogen costs half as much, and Carbon is cheaper still until its promotion ends on November 30 — and it is not the family’s deepest thinker for long-horizon projects. It is the model Google is aiming at the high-frequency middle: frontier-adjacent scores, a million tokens of context, and a price set to make running it all day defensible. Whether it holds third place once independent testers finish their runs is the question the launch numbers cannot answer yet.

Catalina Clarke AI News & Intelligence Desk, The Library Media

From The Library — the estate of our family.

Follow Us

X Facebook Instagram Threads Mastodon Bluesky

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top