Gemini 4 Hydrogen: Google's Fast Gemini 4 Model, Thinking Levels, Pricing, Benchmarks and Availability

What Gemini 4 Hydrogen is

Gemini 4 Hydrogen is the third Gemini 4 model from Google DeepMind, released on 12 November 2026 and announced by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google. Google's model card says it is adapted from Gemini 4 Carbon. Where Carbon and Gemini 4 Argon are built for deep, long-horizon reasoning, Google positions Hydrogen as the fast model for everyday coding and agentic work, used on its own or as the execution model under Carbon or Argon. Google calls it the recommended upgrade from Gemini 3.8 Flash. Its knowledge cutoff is March 2026. On LMArena's text leaderboard Gemini 4 Hydrogen has 1490 Elo (rank 15, high effort).

Where it is available

Unlike Carbon at launch, Hydrogen went to everyone on day one. It is generally available in the Gemini API under the model ID gemini-4-hydrogen, through both GenerateContent and the Interactions API, and in Google AI Studio, on the free tier as well as the paid tier. Google also lists Gemini for business, the Gemini app, Google Search and Google Sheets, and Google Antigravity for paid Google AI Pro and Ultra subscribers. It is now the default model behind the Antigravity agent in Gemini Managed Agents. Google has not published weights, so it cannot be self-hosted.

Context, output and thinking levels

Hydrogen takes text, images, video, audio and PDFs and returns text; it does not generate images or audio and does not work with the Live API. Input, thinking and output share one 1.05M window. A single request returns up to 65,536 tokens, and long decode, through background execution or client-side continuation, carries a response on in 64K-token hops up to the full window. It is the first Gemini 4 model with four thinking levels: low, medium, high and a new xhigh, with high as the default, where Carbon and Argon default to medium and stop at high. The minimal level returns an error, temperature, top_p and top_k are ignored, and explicit context caching is not supported.

Pricing

On the paid tier Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, with thinking tokens billed as output. The price is flat across the whole context window rather than rising for long prompts. Implicitly cached input costs $0.075 per million tokens, a 95% discount, and implicit caching starts on prompts of at least 4,096 tokens. Batch and Flex inference halve the rate to $0.75 input and $3.75 output, and Priority inference costs $2.70 and $13.50. Grounding with Google Search is billed per query after 5,000 free requests a month shared across Gemini 3 and newer models.

Benchmarks, and who ran them

All figures here are Google's own. Its model card reports 57.0% on Terminal-Bench 4.1 after cheating adjustment, against 18.2% for Gemini 3.8 Flash according to the announcement, 13.5% on SWE-sweep, 77.6% on the offline subset of OSWorld-2.1, 71.6% on Agents' Last Exam v1.1, 76.1% on AutomationBench-AA, 53.4% on the Multimodal Knowledge Agent benchmark and 58.8% on HLE-Diamond. Google cites a score of 51 on the Artificial Analysis Intelligence Index at the default high level, one point above Argon, and 54 at xhigh. It also says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash and makes more than five times fewer fabricated claims.

Official Resources