Overview
Gemini 4 Neon is the middle tier of Google DeepMind’s Gemini 4 family. Google released it on November 18, 2026, six days after Gemini 4 Hydrogen, and places it between Hydrogen, the fast everyday model, and Gemini 4 Carbon, the most capable one. According to its model card, Neon is adapted from Carbon. Google aims it at complex coding, analytical reasoning, science and enterprise workflows that run often enough for speed and cost per task to matter, and Neon can also act as an orchestrator that directs parallel Hydrogen subagents in Google Antigravity. Our Gemini 4 Neon launch story covers the release.
Specifications
From Google’s developer documentation and model card:
- Model ID:
gemini-4-neon, a single stable version. - Context: 1,048,576 tokens shared by input, thinking and output. Synchronous responses stop at 65,536 output tokens; background “long decode” requests can run to the full window.
- Input and output: text, images, video, audio and PDF in; text out. No image or audio generation and no Live API.
- Thinking: low, medium, high (default) and xhigh. Carbon and Argon default to medium and stop at high.
- Tools: code execution, computer use, file search, function calling, structured outputs, URL context, and grounding with Google Search and Google Maps.
- Knowledge cutoff: March 2026.
Pricing
| Tier (per 1M tokens) | Input | Output |
|---|---|---|
| Standard | $3.00 | $15.00 |
| Batch and Flex | $1.50 | $7.50 |
| Priority | $5.40 | $27.00 |
| Cached input (Standard) | $0.15 | — |
Prices are flat across the context window, and output includes thinking tokens. Only implicit caching is supported, and there is no free API tier. From December 1, when Carbon and Argon rise to $4 / $20, Neon costs less per token than either.
Availability
Neon is available in the paid Gemini API through Google AI Studio and Android Studio, in Gemini for business, to Google AI Pro and Ultra subscribers in the Gemini app and Google Antigravity, and in Gemini Managed Agents. A cyber-defense version is going to vetted members of Google’s Fairwind Program.
Strengths and limits
The scores in the frontmatter are Google’s own. In its model card Neon leads Claude Opus 5.5 and GPT-6.2 Sol on the Artificial Analysis and Vals indices, but trails Opus 5.5 slightly on Terminal-Bench 4.1 and FrontierCode 2.0. Google says Neon costs 32 percent less per task than Carbon at the default setting, but publishes no tokens-per-second figure. The developer guide warns that hard prompts can take several minutes, that sampling parameters such as temperature are ignored, and that low output limits can cut off answers because thinking tokens count against them. Google lists hallucinations among Neon’s known limits.
Alternatives
Within Google’s lineup, Gemini 4 Hydrogen costs half as much per token, while Gemini 4 Carbon and Gemini 4 Argon are the larger reasoning models. Outside Google, compare Claude Opus 5.5 and GPT-6.2 Sol, covered in our GPT-6.2 launch story.
Specifications are taken from the provider’s official announcement, documentation or model card. Benchmark results depend on evaluation settings; see how AI benchmarks work.