Google Launches Gemini 4 Neon: $3/$15 Mid-Tier Model With Frontier-Level Speed
Google has launched Gemini 4 Neon, a new mid-tier model in the Gemini 4 family, priced at $3 per million input tokens and $15 per million output tokens. Announced on 18 November 2026, Neon sits between Gemini 4 Hydrogen, the fastest model in the family, and Gemini 4 Carbon, the model built for the hardest long-horizon work. Google describes Neon as offering frontier-level intelligence with fast end-to-end task completion, aimed at software engineering, knowledge work and science. It is rolling out now in the Gemini API, and it arrives with benchmark scores that put it at the top of one independent coding-agent ranking.
Where Neon fits in the Gemini 4 family
Gemini 4 is now a four-model family, and each model is aimed at a different trade-off between speed, cost and capability. Hydrogen is the fast option. Carbon is the heavyweight, built for the longest and hardest tasks. Neon fills the middle: a model Google says can handle serious software engineering and knowledge work without the cost or the wait of the top tier. That positioning matters because most real work sits in the middle. Developers and teams do not always need the most powerful model, and they do not always want the cheapest one either. They want a model that is good enough for hard work and fast enough to keep a workflow moving, and that is the gap Neon is built to fill.
Neon's headline claim is speed. Google says it completes tasks about three times faster than Carbon at about 60% lower cost, a vendor claim rather than an independently verified figure. Artificial Analysis, the independent benchmarking site, puts the saving more modestly, noting about 17% lower cost per task than Carbon. The gap between the two figures is a reminder that speed and cost claims depend heavily on what is being measured, and readers should treat both as estimates from interested parties until independent testing catches up.
![]() |
| Credit: Negative Space (via StockSnap) |
Price, context and thinking levels
Neon's pricing is straightforward. Input tokens cost $3 per million, output tokens cost $15 per million, and cached input costs $0.15 per million, which Google calls a 95% discount on the input price. There is no free tier, so every use of Neon is paid. The model has a 1-million-token context window, precisely 1,048,576 tokens, which lets it hold very large codebases, documents or research sets in a single session.
Neon also introduces four thinking levels: low, medium, high and xhigh, with high as the default. A thinking level is a setting that controls how much reasoning the model does before it answers. Lower levels answer faster and cost less, while higher levels spend more compute working through a problem and are meant for harder tasks. At the default high level, Neon can produce up to 65,536 output tokens in a single synchronous response, and up to 1 million tokens through background long decode, a mode that lets the model keep working on a long output in the background. The xhigh level is where Neon posts its best benchmark scores, so users chasing maximum capability will need to accept the extra time and cost that level brings.
Benchmark scores: what Google reports
All of the scores below are Google-reported, some of them through Artificial Analysis, and none has been independently verified by INORYVA. At the xhigh thinking level, Neon scores 61 on the Artificial Analysis Intelligence Index, placing it third overall. On the Artificial Analysis Coding Agent Index, Neon ranks first with a score of 58.6, ahead of the models it is most often compared with. At the default high level, it scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0, two benchmarks that test real coding work rather than short quiz-style questions. Neon also ranks second on the Vals Index.
The coding-agent ranking is the most important of these for the audience Google is targeting. A coding agent is an AI system that does not just suggest a line of code but carries out a whole task, such as fixing a bug or building a feature, across a codebase. Neon's top ranking there, combined with its mid-tier price, is the core of Google's pitch: frontier-level coding ability at a price and speed that make it practical for everyday development work. For context on the model Neon sits below, see our report on Gemini 4 Carbon, and for the wider field, the flagship AI model comparison.
Where you can use Neon today
Neon began rolling out on 18 November in the Gemini API, through AI Studio, in Gemini for business, in Google Antigravity for paid AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra users. It is also available in GitHub Copilot for Pro+, Max, Business and Enterprise plans, in Cursor, and through the Vercel AI Gateway. That distribution matters: Neon is not confined to Google's own tools, and developers who already work in Copilot or Cursor can try it without changing their editor.
Google is also releasing a defensive-security variant called Gemini 4 Neon for cyber, which is rolling out to trusted defenders through the Fairwind Program in the coming days. The variant is aimed at security teams rather than general users, and Google has not published separate pricing or benchmark figures for it. Neon's knowledge cutoff is March 2026, and its model card warns that it can produce fabricated information, a standard caution for large language models that readers should keep in mind when using its output for anything that needs to be factually exact.
![]() |
| Credit: Negative Space (via StockSnap) |
Neon against the competition
Neon's most direct rival is OpenAI's GPT-6.2 Astra, launched the day before Neon. Astra costs $8 per million input tokens and $40 per million output tokens, more than double Neon's price on both measures, and it scores 62.1 on the Artificial Analysis Intelligence Index, one place ahead of Neon. But Neon leads the Coding Agent Index, where Astra's Codex implementation scores 53. The two models therefore win on different measures, and the right choice depends on whether a team values overall intelligence scores or coding-agent performance and price. Our full report on GPT-6.2 Astra sets out its pricing and scores in detail.
The launch also continues a pattern in Google's consumer AI: features that start in the paid tiers spread outward over time. Neon is available in the Gemini app only to Pro and Ultra subscribers at launch, while the free tier has no access. That is a contrast with Gemini Tasks free six countries, where Google opened a feature to free users in selected markets. For now, Neon is a paid product, and its price is the point: Google is betting that a capable mid-tier model at $3 and $15 per million tokens is where most developers will actually spend their money.
![]() |
| Credit: Negative Space (via StockSnap) |
Price and scores at a glance
| Measure | Gemini 4 Neon |
|---|---|
| Input price (per 1M tokens) | $3.00 |
| Output price (per 1M tokens) | $15.00 |
| Cached input price (per 1M tokens) | $0.15 (95% discount, per Google) |
| Free tier | None |
| Context window | 1M tokens (1,048,576) |
| Thinking levels | Low, medium, high (default), xhigh |
| Max sync output tokens | 65,536 (1M via background long decode) |
| Artificial Analysis Intelligence Index | 61 at xhigh (#3 overall) |
| Artificial Analysis Coding Agent Index | 58.6 (#1) |
| Terminal-Bench 4.1 (default high) | 65.2% |
| FrontierCode 2.0 (default high) | 46.2% |
| Vals Index | #2 |
Key Takeaways
- Gemini 4 Neon launched on 18 November 2026, priced at $3 input and $15 output per million tokens, with cached input at $0.15 and no free tier.
- It sits between Hydrogen and Carbon in the Gemini 4 family, aimed at software engineering, knowledge work and science.
- At xhigh it scores 61 on the Artificial Analysis Intelligence Index (#3) and ranks #1 on the Coding Agent Index with 58.6.
- It offers four thinking levels, a 1-million-token context window and up to 65,536 synchronous output tokens.
- It is available now in the Gemini API, AI Studio, Antigravity, the Gemini app for Pro/Ultra, GitHub Copilot, Cursor and Vercel AI Gateway.
Conclusion
Gemini 4 Neon is Google's attempt to own the middle of the AI model market: a model fast and cheap enough for everyday development, but capable enough to top a coding-agent ranking. The $3 and $15 price, the 1-million-token context and the four thinking levels give developers real control over the cost and capability trade-off. The open question is whether the benchmark scores hold up outside Google's own reporting, and whether developers find the mid-tier trade-off better in practice than the cheaper or the more powerful alternatives. The first answers will come from the Copilot, Cursor and API users who can try it today.
Sources
- Google, "Introducing Gemini 4 Neon" (18 November 2026), by Koray Kavukcuoglu: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/



Comments
Post a Comment