Google introduced Gemini 4 Hydrogen on November 12, a fast model for coding and agent work, in its launch post. Google reports that it finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting and scores 57% on Terminal-Bench 4.1 against 18.2% for the older model. It is priced at $1.50 per million input tokens and $7.50 per million output tokens, flat across a 1 million-token window.
Gains over 3.8 Flash cover speed, coding scores and fabrication rates
Besides the Terminal-Bench result, Google says Hydrogen cuts fabricated claims by more than five times against 3.8 Flash. On the Artificial Analysis Intelligence Index it reports 51 at the default high thinking level, against 50 for its frontier Gemini 4 Argon, and 54 at the new xhigh level. Hydrogen has four thinking levels, low, medium, high and xhigh, and high is the default.
Orchestrator in Antigravity mixes Hydrogen with Carbon and uses a different Terminal-Bench version
Google also introduced Orchestrator in Antigravity, which switches between Carbon and Hydrogen in one session. On Terminal-Bench 4.0 it reports near-Carbon success with 24% less quota and 19% faster completion. That is a different benchmark version from the 4.1 used for the 3.8 Flash comparison, so the two sets of figures may not line up.
The 1 million output tokens in the launch post are shared with input and need long decodes
The launch post describes native 1 million-token input with up to 1 million output tokens. Google’s developer guide sets a single limit of 1,048,576 tokens that covers input and output together, with thinking and function calls counted as output. One request returns up to 65,536 output tokens by default, and longer generations need long decodes, which can run up to 1,048,576 tokens minus the input. A long prompt therefore likely shrinks the room left for the answer.
Availability reaches the API, the Gemini app and Search, and Antigravity for paid plans
Hydrogen is offered in Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and in Antigravity for paid Google AI Pro and Ultra subscribers. The benchmark and speed figures come from Google, and the launch post appears not to link independent tests of them.