Google Releases Gemini 4 Hydrogen, a Cheaper Model That Edges Past Its Flagship Argon

On the Artificial Analysis Intelligence Index at its default high thinking level, Hydrogen scored 51 and Argon 50, Google said. A new fourth thinking level, called xhigh, raises Hydrogen's score to 54. Koray Kavukcuoglu, Google DeepMind's senior vice president and Google's chief AI architect, wrote in the announcement that Hydrogen is "our daily driver for agentic workflows". The heavier Gemini 4 Carbon and Argon models are left for long, deep reasoning tasks.

Cost is what matters most here. Hydrogen is priced at $1.50 per million input tokens and $7.50 per million output tokens, the same rate across its full 1 million-token context window. Argon costs $2 and $10 until its introductory pricing ends on November 30, when it rises to $4 and $20, according to SiliconANGLE. For companies running agents at volume, a model that matches the flagship at well under half its standard price changes the economics of production workloads.

Most of Google's comparisons are against Gemini 3.8 Flash, released in September. The company said Hydrogen finishes tasks about 2.5 times faster than that model at the default setting. On the Terminal-Bench 4.1 command-line test it scored 57%, against 18.2% for the older model. Early partners testing graph, SQL and data-heavy enterprise workloads saw runs 2 to 2.8 times faster and token use 12% to 23% lower than with 3.8 Flash, Google said. It also reported scores of 52.3% on the Vals Tax Agent Bench and 55.8% on the Vals Legal Research Bench. All these figures come from Google.

The launch is quicker and wider than Argon's. That model spent nearly two weeks with members of Google's Fairwind Program, a group of vetted cybersecurity defenders, before paying developers could use it from October 13, SiliconANGLE reported. Hydrogen skipped that stage. It is available now in the Gemini API, Google AI Studio, Android Studio, Gemini for business, the Gemini app's model menu and Google Search. Free Gemini app users get Hydrogen through automatic model selection when it suits a prompt, though Google said most prompts there still go to Flash-Lite. Paid Google AI Pro and Ultra subscribers can use it in Gemini in Google Sheets and in Antigravity, Google's coding tool.

Antigravity also gained a feature called Orchestrator, which switches between Carbon and Hydrogen during a session. On Terminal-Bench 4.0, Google said, Orchestrator came close to Carbon's success rate while using about 24% less quota and finishing tasks about 19% faster on average.

Google also made safety claims for agent use. It said Hydrogen reduces overreach on agentic tasks by up to 81%, cuts misaligned actions and task shortcutting across tool-use and coding evaluations twelvefold, and is up to twice as unlikely to get around permission boundaries when a task is blocked. It said the model is more than 2.3 times as resilient on the Gray Swan indirect prompt injection benchmark.

According to SiliconANGLE, the early testers Google quoted included Andrew Carr, chief scientist at 3D animation startup Cartwheel, and Discovery Loop, the startup co-founded by former Google chief scientist Jeff Dean after he left in August.

Hydrogen is the third Gemini 4 model to ship in just over six weeks. It came two days after Carbon opened to paying developers and Google AI Ultra subscribers. Argon's lower introductory price ends on November 30.