Introducing Gemini 4 Hydrogen

Open original

Google has released Gemini 4 Hydrogen, a faster, cheaper model meant for everyday coding, agentic workflows, and consumer tools, sitting below its frontier models Carbon and Argon. The pitch is that it matches or beats its class on many benchmarks while finishing tasks quicker and costing $1.50 per million input tokens.

  • Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, edging past Argon's 50, and reaches 54 at its highest thinking level.
  • Google says it finishes tasks 2.5 times faster than Gemini 3.8 Flash and more than triples its Terminal-Bench 4.1 score, from 18.2% to 57%.
  • A new Orchestrator in Google Antigravity switches between Carbon and Hydrogen in one session, reportedly using 24% less quota and finishing 19% faster than Carbon alone.

Google is releasing Gemini 4 Hydrogen, a faster and cheaper model sitting below its frontier models Carbon and Argon. Hydrogen is aimed at everyday coding, agentic tasks, and consumer use, and Google is rolling it out broadly: Google AI Studio across all tiers, the Gemini API, Gemini for business, the Gemini app, Google Search, and paid AI Pro and Ultra subscribers in Google Antigravity. Pricing is $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1 million-token context window. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level, edging Argon's 50, and reaches 54 at a new xhigh setting. Google says it usually finishes tasks faster than comparable models. The more interesting pitch is how Hydrogen fits alongside the bigger models. Google introduced an Orchestrator in Antigravity that switches between Carbon and Hydrogen inside a single session, with Carbon handling architectural judgment and Hydrogen doing the grunt work of writing code, running tests, and checking changes. On Terminal-Bench 4.0, Google reports Orchestrator reaching near-Carbon success rates while using about 24% less quota and finishing 19% faster. Parallel setups work similarly, with Carbon coordinating and Hydrogen running worker and verifier agents in isolated workspaces. Against its predecessor, Gemini 3.8 Flash, Google claims Hydrogen finishes tasks 2.5 times faster at default settings, more than triples the Terminal-Bench 4.1 score (57% versus 18.2%), and cuts fabricated claims by more than fivefold. Partners testing complex graph, SQL, and data-heavy workflows reported 2x to 2.8x speedups and 12% to 23% lower token use. Google's benchmark claims are the part to take with some salt, since they are self-reported and mostly from Google's own selection of tests. The numbers that stand out include 58.8% on HLE-Diamond, a #1 rank on Vals Tax Agent Bench at 52.3%, and 71.6% on Agents' Last Exam v1.1, though Hydrogen only ranks #5 on Agent Arena for open-ended agentic work. The article also says Hydrogen reduces overreach in agentic tasks by up to 81%, which sounds meaningful but is cut off in the copy provided, so the rest of the safety section isn't visible here. For someone deciding whether to read the full piece, the useful parts are the pricing, the multi-model orchestration design, and the Terminal-Bench figures, which are concrete enough to test against your own workloads.

Ask about this article