Google rolls out Gemini 4 Hydrogen across AI products

Google began rolling out Gemini 4 Hydrogen on November 12, putting the model into AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search. Standard API access costs $1.50 per million input tokens and $7.50 per million output tokens.

The rollout also covers Google Antigravity for paid AI Pro and Ultra subscribers, Android Studio and eligible Gemini features in Google Sheets. Google describes Hydrogen as the rapid-execution member of its Gemini 4 generation, alongside the deeper-reasoning Gemini 4 Carbon and Argon models.

Google said the global rollout started on November 12, but it did not give a date when every announced account, region and product surface would have access.

For developers, Google lists Gemini API pricing of $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens. Cached input costs $0.075 per million tokens. Batch and Flex access each cost $0.75 per million input tokens and $3.75 per million output tokens. The invite-only Contributor tier carries the same token rates.

Google’s developer guide presents gemini-4-hydrogen as a migration target for Gemini 3.8 Flash and earlier Flash models. The migration requires a new model identifier and changes the default thinking level from MEDIUM to HIGH, with a new XHIGH option. The guide also tells developers to remove unsupported thinking_budget, MINIMAL thinking-level and explicit caching settings.

The migration guidance makes Hydrogen the successor path for Flash workloads, but the primary sources reviewed for this story do not set a shutdown date for Gemini 3.8 Flash.

Google says Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash at Hydrogen’s default HIGH setting. The company also reports a 57% score for Hydrogen, compared with 18.2% for Flash, in its Terminal-Bench 4.1 comparison and says the model reduces fabricated claims by more than fivefold. Those performance and reliability figures are Google-reported and were not independently reproduced in the evidence reviewed for this story.

Google documents a 1,048,576-token combined input-output limit for Hydrogen, with text, image, audio and video inputs. Ordinary single-segment foreground output is capped at 65,536 tokens. Google’s model card says Hydrogen can hallucinate, may respond slowly or time out, can consume more tokens at higher thinking levels and has a March 2026 knowledge cutoff.