Google releases Gemini 4 Hydrogen for its API, Gemini app and Search
Hydrogen, described by outlets as the successor to Gemini 3.8 Flash, is rolling out in Google's API, app and Search. Independent tests show mixed results.
Key takeaways
- Google released Gemini 4 Hydrogen on Nov. 12 and is rolling it out worldwide in the Gemini API, the Gemini app and Google Search.
- Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting and more than tripled its Terminal-Bench 4.1 score.
- API pricing is $1.50 for each million input tokens and $7.50 for each million output tokens, with a 1 million-token context window.
- Artificial Analysis scores Hydrogen at 54 on its Intelligence Index at xhigh but finds it slower per task than average and very verbose.
- The Decoder reports Hydrogen trails Claude Sonnet 5.5 on Terminal-Bench 4.0 in Artificial Analysis testing, while scoring low on fabrication.
Google on Nov. 12 released Gemini 4 Hydrogen, a model it positions for quick everyday work such as coding and agent tasks, and began rolling it out worldwide in the Gemini API, the Gemini app and Google Search. In its announcement, Google said Hydrogen is meant for fast execution, while its larger Gemini 4 Carbon and Gemini 4 Argon models handle long, difficult reasoning. The Decoder and Silicon Report describe Hydrogen as the replacement for Gemini 3.8 Flash; Google itself pitches it as a major upgrade for teams moving off that model.
What does Google say Hydrogen can do?
Google says that, at its default "high" thinking setting, Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash. It also says the model more than tripled the older model's Terminal-Bench 4.1 score (57% vs. 18.2%) and cut fabricated claims by more than 5x. Users can pick from four thinking levels, with xhigh added above low, medium and high. According to Google, it takes up to 1 million tokens of input and can produce up to 1 million output tokens.
Google also claims that, at the default setting, Hydrogen outscores Gemini 4 Argon on the Artificial Analysis Intelligence Index (51 vs. 50), reaching 54 at xhigh. The Decoder argues that this leaves Argon looking obsolete less than four weeks after its launch.
In Google Antigravity, its coding tool, Google is adding a feature called Orchestrator that switches between Carbon and Hydrogen within one session. On Terminal-Bench 4.0, Google says, Orchestrator neared Carbon's task success rates while consuming 24% less quota and completing work 19% faster on average.
What does it cost, and who gets it?
Google charges $1.50 for each million input tokens and $7.50 for each million output tokens, with the same rate across the full context window. The Decoder reports that cache hits cost $0.075. Developers can use it through the Gemini API, in Google AI Studio and in Android Studio. Antigravity access is open to paying Google AI Pro and Ultra subscribers. Silicon Report says free users of the Gemini app are also routed to Hydrogen automatically when the app decides a prompt suits it.
In Search, Google calls Hydrogen "the biggest model upgrade to Search yet." Robby Stein, Google's VP of Product for Search, wrote that it is available to all users worldwide and is meant to handle multi-part questions and typos better, show more visual answers such as comparison tables, and link to a broader range of sites.
What do independent tests show?
Artificial Analysis, which runs its own benchmarks, gives Hydrogen at xhigh a score of 54 on its Intelligence Index, against a median of 41 for comparable models. Its listing also rates the model as slower per task than average and "very verbose," noting that it generated 304M tokens during the evaluation compared with a median of 72.8M. According to The Decoder, Artificial Analysis puts the cost per task at $3.16 at xhigh and $1.87 at high.
The Decoder reports that, in the Artificial Analysis results, Hydrogen at its highest setting scores 54, one point behind Anthropic's Claude Sonnet 5.5 and OpenAI's GPT-6.1 Sol, both of which reach 55 at their top settings. The same results show it making things up only 8 percent of the time on the AA-Omniscience test, but scoring 54 percent on Terminal-Bench 4.0, below Claude Sonnet 5.5 at 64 percent and GPT-6.1 Sol at 56 percent. Google's own comparison, as reported by Silicon Report, puts Hydrogen ahead of GPT-6.1 Terra and Claude Sonnet 5.5 on knowledge-work and multimodal tests, but behind both on science and math, computer use and Terminal-Bench 4.1, gaps Google described as small.
What about safety?
According to Google, Hydrogen cuts agentic overreach by up to 81%. When a task is blocked, Google says, the model is up to 2x less likely to get around permission limits. It also reports a 94% honesty rate on its own internal agent honesty evaluation. These figures come from Google's own testing, and NextFish has not independently verified them. Google points readers to the model card for more detail.
Correction: An earlier version of this story gave a higher honesty rate, matching Google's original post. Google's post now gives the figure as 94%.
Disclosure: NextFish's AI drafting desk uses Claude, a model made by Anthropic.
Questions readers are asking
What is Gemini 4 Hydrogen?
A Google model for fast everyday work such as coding and agent tasks. Google positions it alongside its larger Carbon and Argon models, and outlets describe it as the replacement for Gemini 3.8 Flash.
How much does Gemini 4 Hydrogen cost?
Google charges $1.50 for each million input tokens and $7.50 for each million output tokens. The Decoder reports cache hits cost $0.075.
Is Hydrogen available in Google Search?
Yes. Google says it is now used in Search for all users worldwide.
How does it compare with rival models?
According to The Decoder, Artificial Analysis scores it one point behind Claude Sonnet 5.5 and GPT-6.1 Sol at their highest settings, and behind both on Terminal-Bench 4.0. Google's own chart shows it ahead of Sonnet 5.5 and GPT-6.1 Terra on knowledge-work and multimodal tests.
Sources
- Google: Introducing Gemini 4 Hydrogen
- Google: Google Search gets a major upgrade with Gemini 4 Hydrogen
- The Decoder: Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch
- Artificial Analysis: Gemini 4 Hydrogen (Xhigh) Intelligence, Performance & Price Analysis
- Silicon Report: Google launches Gemini 4 Hydrogen, replacing Gemini 3.8 Flash
- #gemini
- #ai-models
- #google-search
- #benchmarks
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
AI2 min
HUMAIN sets Dec. 15 for its AI stack, promises a 3T-parameter model
The Saudi AI company's CEO calls the unnamed model the world's largest open-weight model and says it is far cheaper to run. No model card, licence or benchmark has been published.
Agents3 min
Google opens Antigravity's /teamwork agents, claims math and code results
Google Research says its multi-agent system helped prove a 2013 conjecture and port large codebases to Rust, and offers 10,000 academics free access. Results are Google's own.
Cybersecurity3 min
Google Project Zero cuts bug disclosure deadline to 30 days by 2028
The Google team says AI has sped up attacks. It drops the extra 30-day patch soak time now and will cut vendors' deadline from 90 to 60, then 30 days.
Big Tech2 min
Google lets Gemini read Google Wallet data to answer money questions
Google is rolling out a Wallet connection for US Gemini users, and optionally bank and card accounts via Plaid, for budgeting help. Google says access is read-only and opt-in.