Google clarifies Gemini 4 Hydrogen parameters: 2.5× faster than Gemini 3.8 Flash, fewer fabricated claims
Google has added specific parameters for the Gemini 4 Hydrogen model: according to the company, it is 2.5× faster than Gemini 3.8 Flash, reduces fabricated claims by more than 5×, and on the Artificial Analysis Intelligence Index benchmark it even beats the more powerful Gemini 4 Argon model. The model replaced Gemini 3.8 Flash in products…
Google has released further details on the Gemini 4 Hydrogen model, which recently replaced Gemini 3.8 Flash. According to the company, the model, even in its high reasoning mode, is 2.5× faster than Gemini 3.8 Flash and reduces the number of "fabricated claims" by more than fivefold. In code generation, according to Ars Technica, Hydrogen is said to more than triple its predecessor's score on the Terminal-Bench benchmark. On the Artificial Analysis Intelligence Index benchmark, Hydrogen scored 54, surpassing the more powerful Gemini 4 Argon model, which scored 50 — however, sources disagree on which reasoning mode was used to achieve this result: Ars Technica states the "xhigh" mode, while TechCrunch states the default "high" mode. On multi-step tasks such as deep research or data analysis, the model ranked 5th on the Agent Arena leaderboard, according to TechCrunch.
The model offers four reasoning levels (low, medium, high, xhigh) and is primarily intended for agentic workflows — tasks in which the model independently performs multiple steps, including writing code, running tests, or verifying code changes. It can work on its own or in combination with the more powerful Gemini 4 Carbon and Gemini 4 Argon models.
Gemini 4 Hydrogen is available globally to all Google Search users, where it provides better understanding of complex and multi-part queries, more accurate discovery of related websites thanks to an improved fan-out technique, and more visual answers (images, tables, structured overviews) in more languages. Outside Search, the model has been deployed across all tiers of Google AI Studio, in the Gemini API, in the Gemini app, and for AI Pro and Ultra subscribers on the Antigravity platform, without the gradual access rollout seen with the previous Argon and Carbon models.
Why it matters
For ordinary Google Search users, this means a direct improvement in the quality of answers to complex, multi-part, or imprecisely formulated queries, including visual overviews and better support for multiple languages. For developers and companies working with the Gemini API or AI Studio, Hydrogen offers a cheaper and faster alternative to more powerful models for agentic tasks and code generation, without having to wait for the gradual access rollout seen with earlier models.
Release card
Gemini 4 Hydrogen
- Availability
- Available globally in Google Search for all users, as well as in all tiers of Google AI Studio, in the Gemini API, in the Gemini app, and for AI Pro and Ultra subscribers on the Antigravity platform, without gradual access restrictions.
- Artificial Analysis Intelligence Index 54 bodů (model Gemini 4 Argon dosáhl 50 bodů) In this general intelligence test, the model achieved a higher score than the more powerful Gemini 4 Argon model; sources differ on whether this was in the xhigh reasoning mode or the default high mode.
- Agent Arena 5. místo On multi-step tasks such as deep research, data analysis, or codebase navigation, the model ranked fifth in this agentic benchmark leaderboard.
- Rychlost dokončení úloh oproti modelu Gemini 3.8 Flash 2,5× rychlejší i v režimu uvažování high According to Google, the model completes tasks faster than its predecessor Gemini 3.8 Flash, even at the high reasoning level.
- Vymyšlená tvrzení (fabricated claims) pokles o více než 5× oproti modelu Gemini 3.8 Flash According to Google's analysis, the model produces significantly fewer fabricated claims than its predecessor.
- Agentic tasks such as writing code, running tests, and verifying code changes, either on its own or in combination with the Carbon and Argon models
- Searching in Google Search with complex, multi-part queries and imprecise input (typos)
- Multi-step tasks such as deep research, data analysis, or codebase navigation
- According to TechCrunch, it is not intended for the deepest reasoning on difficult problems, where, according to Google, the Gemini 4 Carbon and Gemini 4 Argon models are better
The model replaced Gemini 3.8 Flash and, according to Google, is designed for fast agentic tasks, while the more powerful Gemini 4 Carbon and Gemini 4 Argon models remain intended for deep reasoning on difficult problems.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
What was added since the original report
Verified updates
-
The model is 2.5× faster than Gemini 3.8 Flash; Primarily intended for agentic workflows and multi-step tasks; Offers four reasoning levels (low, medium, high, xhigh); Available in the Gemini API, AI Studio, Gemini App, AI Pro and Ultra, not just in Search; Surpasses Gemini 4 Argon on the Artificial Analysis Intelligence Index benchmark
- The model is 2.5× faster than Gemini 3.8 Flash
- Primarily intended for agentic workflows and multi-step tasks
- Offers four reasoning levels (low, medium, high, xhigh)
- Available in the Gemini API, AI Studio, Gemini App, AI Pro and Ultra, not just in Search
- Surpasses Gemini 4 Argon on the Artificial Analysis Intelligence Index benchmark
-
Gemini 4 Hydrogen immediately replaces Gemini 3.8 Flash in Google's products; The model reduces fabricated claims by more than 5× compared to Flash 3.8; The model achieves a score of 54 on the Artificial Analysis Intelligence Index in xhigh mode; Rollout is immediate for all users without gradual restriction; The model is named after an element (Hydrogen) in the new naming scheme (Argon, Carbon)
- Gemini 4 Hydrogen immediately replaces Gemini 3.8 Flash in Google's products
- The model reduces fabricated claims by more than 5× compared to Flash 3.8
- The model achieves a score of 54 on the Artificial Analysis Intelligence Index in xhigh mode
- Rollout is immediate for all users without gradual restriction
- The model is named after an element (Hydrogen) in the new naming scheme (Argon, Carbon)
Two audiences, two different impacts
What this means
For individuals
Google Search users get more accurate answers to complex and multi-part queries, including visual overviews and better language support, without having to change anything.
For a business
Companies using the Gemini API or AI Studio now have access to a faster and, according to the company, cheaper model for agentic workflows and code generation, which can replace more expensive models in tasks where maximum performance is not required.
DevelopmentCheck the original
Event sources
confirmed by 2 independent sources · 3 publishers, 2 independent. We count feeds from the same owner only once.