Coding is Love
News 4 min read

Google releases Gemini 4 Hydrogen, its fast model, in Search and API

Google released Gemini 4 Hydrogen on November 12, 2026, a faster and cheaper model for everyday coding and agent work, and made it the model behind Google Search on the same day. In the announcement by Koray Kavukcuoglu, Google says “Hydrogen brings cutting-edge intelligence to everyone.” Developers can call it today in the Gemini API and AI Studio on all tiers, at $1.50 per million input tokens.

What’s new

Hydrogen joins the two frontier models Google already ships, Gemini 4 Argon and the Gemini 4 Carbon model released two days ago. Google positions Carbon and Argon for deep, long-running reasoning, and Hydrogen for quick execution: it can run on its own or act as the execution engine next to Carbon and Argon in a larger agent setup.

Context and pricing. Hydrogen takes up to 1 million input tokens and can return up to 1 million output tokens. It costs $1.50 per million input tokens and $7.50 per million output tokens, and Google says that price is flat across the whole 1M window, with no higher rate for long prompts.

Thinking levels. There are four: low, medium, high (the default) and a new xhigh level for harder tasks.

Benchmarks. These are Google’s own numbers:

Benchmark Gemini 4 Hydrogen
Artificial Analysis Intelligence Index 51 at high, 54 at xhigh (Argon: 50)
Terminal-Bench 4.1 57% (Gemini 3.8 Flash: 18.2%)
HLE-Diamond 58.8%
Agents’ Last Exam v1.1 71.6%
Multimodal Knowledge Agent 53.4%
Vals Tax Agent Bench 52.3%, first overall
Vals Legal Research Bench 55.8%, tied for first

Google also lists it fifth on the Agent Arena leaderboard at arena.ai.

Speed and cost against older models. Google says Hydrogen is about 2.5x faster than Gemini 3.8 Flash at the default high setting, and that early partners saw it run 2x to 2.8x faster while using 12% to 23% fewer tokens. Google also introduced Orchestrator in Google Antigravity, which switches between Carbon and Hydrogen in one session; on Terminal-Bench 4.0 it reached near-Carbon task success while using about 24% less quota and finishing about 19% faster on average.

Safety. Google reports that fabricated claims dropped by more than 5x against Gemini 3.8 Flash, misaligned actions and task shortcutting fell 12x in tool-use and coding evaluations, and the model scored a 96.7% honesty rate on an internal test. It also says Hydrogen is more than 2.3x more resilient on the Gray Swan indirect prompt injection benchmark. A model card is linked from the announcement.

In a separate post, Robby Stein, VP of Product for Google Search, calls it “the biggest model upgrade to Search yet.” Hydrogen is rolling out worldwide to all Search users. Google says Search now copes better with long, multi-part questions, typos and very specific requests, runs more parallel searches on related subtopics, links to a wider range of websites, and shows more images, comparison tables, structured layouts and custom tools in its answers. The better responses apply in every language AI Mode supports.

Where you can use it

Who Where
Developers Gemini API and Google AI Studio (all tiers), Android Studio (paid tier), Google Antigravity for Google AI Pro and Ultra subscribers
Businesses Gemini for business (API or model menu), Google Antigravity, Gemini in Google Sheets for eligible plans
Everyone Gemini app model menu (free users get it through automatic model selection), Google Search, Gemini in Google Sheets for AI Pro and Ultra

Google says the rollout is global starting today. The announcement does not mention Vertex AI.

What it means for developers

The model ID in AI Studio is gemini-4-hydrogen, so trying it is a one-line change if you already call Gemini through the API. The free tier includes it, which is not the case for Carbon, still limited to the paid tier.

The pricing is the part to look at first. At $1.50 in and $7.50 out per million tokens with no long-context surcharge, Hydrogen is cheaper than Argon and Carbon ($2 in and $10 out per million during Carbon’s launch offer), and the Artificial Analysis Index score Google reports puts it level with or above Argon. For agents that make many short tool calls, such as code edits, test runs or data lookups, Google’s suggested pattern is to let Carbon or Argon plan and Hydrogen carry out the steps.

Two cautions. All the benchmark and speed numbers above come from Google and its launch partners; run your own tasks before switching a production workload. And the xhigh level uses more thinking tokens, so measure the cost per task, not just the per-token price.

For site owners, Search fanning out to more subtopics and linking to “a wider range of websites” may change which pages get cited in AI answers. Watch your Search Console traffic over the next few weeks.

How it compares

  • Against Gemini 3.8 Flash, the earlier fast model Google measures it against: about 2.5x faster, and 57% on Terminal-Bench 4.1 against 18.2%.
  • Against Gemini 4 Argon: 51 on the Artificial Analysis Intelligence Index at the default setting, against Argon’s 50, at a lower price.
  • Against Gemini 4 Carbon: Carbon remains Google’s strongest model for long, hard tasks; in Antigravity, Orchestrator, which mixes Carbon and Hydrogen, reached near-Carbon success with about a quarter less quota.

Sources