Introducing Gemini 4 Hydrogen

Open original

Google has released Gemini 4 Hydrogen, a faster and cheaper model built for coding, agent workflows, and everyday tasks, rolling out across the Gemini app, Search, AI Studio, and enterprise tools. It sits alongside the heavier Carbon and Argon models, which handle deeper reasoning.

  • Hydrogen scores 51 on the Artificial Analysis Intelligence Index at default settings, just above Argon's 50, and finishes tasks faster than comparable models. It costs $1.50 per million input tokens and $7.50 per million output tokens.
  • Against its predecessor Gemini 3.8 Flash, Google reports tasks finishing 2.5 times faster, Terminal-Bench 4.1 scores rising from 18.2% to 57%, and fabricated claims cut by more than 5x.
  • The Orchestrator feature in Google Antigravity switches between Carbon and Hydrogen within a session, which Google says reached near-Carbon task success while using 24% less quota and running 19% faster.

Google's new Gemini 4 Hydrogen is a fast, cheaper model meant for day-to-day execution work, sitting beneath the company's frontier models, Gemini 4 Carbon and Gemini 4 Argon, which handle deep, long-horizon reasoning. Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, edging out Argon's 50, and reaches 54 at its top xhigh setting. It is priced at $1.50 per million input tokens and $7.50 per million output tokens, flat across a 1 million-token context window. Google is rolling it out broadly, including to free-tier users of AI Studio and the Gemini API, Gemini for business, the Gemini app, and Google Search, with paid Google AI Pro and Ultra subscribers getting it in Google Antigravity. The more interesting pitch is the multi-model setup. Google introduced an Orchestrator in Antigravity that switches between Carbon and Hydrogen within a single session, so the expensive model handles architectural judgment and the cheap one does the grinding work of writing code, running tests, and verifying changes. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon success rates while using 24% less quota and finishing 19% faster. For parallel agent work, Carbon can coordinate while Hydrogen runs worker and verifier agents across isolated workspaces. For teams moving off Gemini 3.8 Flash, the claimed gains are large: tasks finish 2.5 times faster at the default setting, Terminal-Bench 4.1 scores rise from 18.2% to 57%, and fabricated claims drop by more than fivefold. Early partner tests on SQL and data-heavy workflows showed 2x to 2.8x speedups and 12% to 23% lower token use. Keep in mind these are Google's own numbers and partner reports, not independent results. On the consumer side, the changes are mostly about instruction-following and fewer back-and-forth turns: tighter adherence to tone and format, better tool calling across Gmail, Drive, and connected apps like Airtable and Dropbox, and Search that handles messy, multi-part queries and returns more visual answers and a wider spread of links. Google also claims cuts of up to 81% in agentic overreach and 79% in under-service, though the article is cut off before it explains those figures. Hydrogen's four thinking levels (low, medium, high, and the new xhigh) let users trade latency and cost against reasoning depth, which is a sensible design if you actually use it, since the default high setting is where most people will live.

Ask about this article