Model profile & tracker
Gemini 4 Hydrogen: Price, Context and Specs (Updated)
Gemini 4 Hydrogen (GA 12 November 2026): $1.50/$7.50 per million tokens, 1.05M-token context, 95% cache discount, free tier. Specs and vs Neon and Qwen4-Max.
Updated 22 November 2026. This page tracks Gemini 4 Hydrogen — Google's everyday model made generally available on 12 November 2026 — and gets rewritten when pricing, tiers or context limits change.
Gemini 4 Hydrogen is the cheapest way to put a million tokens of multimodal context in front of a Google model: $1.50 input / $7.50 output per million tokens, flat across the full 1,048,576-token window, with text, image, video, audio and PDF input and a rate-limited free tier. Cached input costs $0.075 per million — a 95% implicit-cache discount.
The specs, as published
| Spec | Value |
|---|---|
| Model ID | gemini-4-hydrogen |
| Generally available | 12 November 2026 |
| Context window | 1,048,576 tokens (unified input/output) |
| Default max output | 65,536 tokens |
| Max output (long decode) | Up to 1,048,576 tokens |
| Inputs | Text, images, video, audio, PDF |
| Output | Text |
| Input price | $1.50 / 1M tokens |
| Output price | $7.50 / 1M tokens (thinking tokens included) |
| Cached input | $0.075 / 1M tokens (95% discount) |
| Free tier | Yes, rate-limited |
| Knowledge cutoff | March 2026 |
The output price includes thinking tokens: you pay $7.50 per million whether those tokens are reasoning or answer. Adjustable thinking levels — low, medium, high (default) and xhigh — let you cap what a task actually needs, which is the main lever on the bill.
Where it sits in the Gemini 4 family
| Model | Input $/1M | Output $/1M | Position |
|---|---|---|---|
| Gemini 4 Neon | $3.00 | $15.00 | Frontier default, agentic coding and long output |
| Gemini 4 Carbon | — | — | Google's strongest model |
| Gemini 4 Hydrogen | $1.50 | $7.50 | Everyday workhorse, coding and agentic tasks |
Hydrogen is half of Neon's output price for the same 1.05M window. On Google's own numbers it surpasses Gemini 4 Argon on the Artificial Analysis Intelligence Index at its default high thinking level (51 vs 50, reaching 54 at xhigh) while completing tasks faster. Neon keeps the edge on frontier work; Hydrogen is the one you run all day.
The caching story
The $0.075 cached-input price is the real headline for repeat workloads. A retrieval product that sends the same 500K-token corpus on every call pays $750 per million calls for that corpus on fresh input, or $37.50 with implicit caching — a 95% discount applied automatically to reused prefixes, no explicit cache management required. Google's positioning is explicit: make it cheap to keep a large body of context in front of the model.
How to access it
Google AI Studio and the Gemini API, on both the free (rate-limited) and paid tiers, across GenerateContent and the Interactions API. It ships alongside Carbon and Neon with the same unified 1M window and long decode.
Gemini 4 Hydrogen vs the cheap flagships
vs Qwen4-Max ($1.50 / $4.50): same input price, and Qwen's output is 40% cheaper. Hydrogen takes five input modalities to Qwen's two, and offers the 95% implicit-cache discount. If you write much more than you read, Qwen wins the bill; if your corpus is video, audio or PDF, Hydrogen is the only one of the two that can read it. Specs on the Qwen4-Max page.
vs Gemini 4 Neon ($3.00 / $15.00): same window, half the price. Choose Hydrogen unless you specifically need Neon's frontier-tier positioning.
vs Pixtral Large 2502 ($2.00 / $6.00): Pixtral is cheaper on output with image input, at 128K context — a document-and-screenshot model. Hydrogen is the one for full multimodal, million-token work.
FAQ
Is there a free tier? Yes. Hydrogen is on the free tier of the Gemini API and in Google AI Studio, rate-limited, at no charge.
What is the context window? 1,048,576 tokens, unified: input, thinking and output share one window. Default output is capped at 65,536 tokens, extending to the full window with long decode.
Does the output price include reasoning tokens? Yes. Output — including thinking — is $7.50 per million tokens, which is why the adjustable thinking levels matter for cost.
When was it released? Generally available on 12 November 2026. It is the newest row in the comparison table and the model index.