Google Ships Gemini 4 Hydrogen, Claims 2.5x Speed and a Search Upgrade

Google DeepMind released Gemini 4 Hydrogen on Thursday, a fast model that replaces Gemini 3.8 Flash and now powers Google Search for all users globally. Google claims Hydrogen finishes tasks 2.5 times faster than 3.8 Flash at its default setting. Every performance figure in the announcement comes from Google's own testing, not from independent verification.

The pricing is concrete. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, across a 1 million-token context window. Ars Technica reported the price is unchanged from 3.8 Flash. Faster at the same price is a real claim. Independent confirmation is not yet available.

Google positions its frontier models, Carbon and Argon, for long and hard reasoning. Hydrogen targets what Google calls "rapid execution": writing code, running tests and checking changes, alone or as the worker model in multi-agent setups.

What do the benchmarks show?

The numbers are Google's own results, and they are stark.

• Artificial Analysis Intelligence Index: 51 at default settings, 54 at the new xhigh tier, against 50 for the larger Gemini 4 Argon.
• Terminal-Bench 4.1: 57%, against 18.2% for 3.8 Flash.
• HLE-Diamond: 58.8%, which Google says leads its class.
• Agents' Last Exam v1.1: 71.6%, which Google says tops other models.
• Vals Tax Agent Bench: first place at 52.3%.

A smaller model scoring above a frontier model is the notable claim. Google offers no third-party replication. Arena.ai's Agent Arena, at least, is an external ranking; Hydrogen sits fifth there.

Google also says Hydrogen cuts fabricated claims by more than five times compared with 3.8 Flash. Early partners, unnamed in the announcement, reportedly saw runs 2 to 2.8 times faster on data-heavy enterprise workflows while using 12% to 23% fewer tokens. The model has four thinking levels: low, medium, high as default, and a new xhigh tier. It accepts up to 1 million input tokens and returns up to 1 million output tokens.

On safety, Google claims a reduction in overreach on agent tasks of up to 81%, resilience on the Gray Swan indirect prompt injection benchmark improved more than 2.3 times, and false refusals on HealthBench Pro held under 0.4%. These are self-reported figures. The source gives no methodology.

What is Orchestrator?

Alongside Hydrogen, Google introduced Orchestrator in its Antigravity coding environment. It switches between models such as Carbon and Hydrogen within one session. On Terminal-Bench 4.0, Google says Orchestrator came close to Carbon's success rate while using 24% less quota and finishing 19% faster on average. In parallel-agent setups, Carbon coordinates while Hydrogen runs worker and verifier agents.

The rollout is broader than for the frontier models. Ars Technica noted that Carbon and Argon began with restricted access for cybersecurity researchers and then the top subscription tier. Hydrogen became available immediately. Google opened Carbon to all paying API developers on Nov. 10 at an introductory $2 per million input and $10 per million output tokens, rising to $4 and $20 after Nov. 30.

What changes for Search users?

Google calls Hydrogen its "biggest model improvement to Search yet." The company says it handles multipart questions, very specific requests and typos better. It widens the "query fan-out" that splits a prompt into sub-searches, so answers link to more sites. It also brings more images, comparison tables and custom tools to AI Mode responses.

Availability is wide but not uniform. Developers get Hydrogen now in the Gemini API, AI Studio and Android Studio. Businesses access it through Gemini for business and Gemini in Google Sheets. Google AI Pro and Ultra subscribers receive it in Antigravity. In the Gemini app, Hydrogen appears in the model menu, and Google says free users get it through automatic model selection when it fits the prompt. Ars Technica reported that AI Plus and free users still default to an older model in the app. That gap sits quietly beneath the word "globally."

The release followed a Business Insider report on Wednesday that Google was testing a fast model codenamed Hydrogen with staff. The next test is whether independent benchmarks confirm the 2.5x speed claim, and whether unchanged token prices hold as Google pushes Hydrogen into Search at planetary scale.