Google announces Gemini 4 Hydrogen with “significant advances”

Following the availability of Carbon this week, Google today unveiled Gemini 4 Hydrogen instead of keeping the “Flash” branding. This model is presented as bringing “cutting edge intelligence to all”.

Landmarks

Compared to Gemini 3.8 Flash (launched in September), Hydrogen is billed as a “major standalone upgrade” with “significant advancements in speed and intelligence.” It offers a native input of one million tokens and up to one million output tokens. Specifically:

• Hydrogen completes “jobs 2.5 times faster on its high default setting”
• “…more than (triple) the scores of Terminal-Bench 4.1 (57% against 18.2%)”
• Reduces “fabricated allegations by more than 5 times”
• Adapts “cleanly when a tool call fails or a search returns empty”

In terms of benchmarks:

“Hydrogen outperforms Argon on the Artificial Analytical Intelligence Index at its default High Reflection level (51 versus 50, and reaching 54 at its highest x-High Reflection level), while often completing tasks faster than models in its class.”

• “…our partners have seen Hydrogen run 2-2.8x faster while reducing token usage by 12-23%, compared to 3.8 Flash.”
• “For multi-step tasks such as deep research, data analysis, and navigating a large code base, Hydrogen ranks 5th on Agent Arena, arena.ai’s live benchmark. »
• “For in-depth domain reasoning and analysis, it tops its class on HLE-Diamond (58.8%), ranks #1 overall on Vals Tax Agent Bench (52.3%), is tied for first on the Vals Legal Research Bench (55.8%), and leads its class on AutomationBench-AA.
• “It outperforms other models on the new Multimodal Knowledge Agent (MKA) benchmark (53.4%) for end-to-end agentic knowledge work on complex visual files, and on Agents’ Last Exam v1.1 (71.6%) for end-to-end professional workflows.
• Compared to Flash 3.8, Hydrogen “reduces agent task overreach by up to 81% and underservice by up to 79%, and reduces misaligned actions and task shortcuts by 12x when using coding tools and assessments. When faced with impossible or blocked tasks, it is up to 2x less likely to bypass permission limits and achieves a 96.7% honesty rating on our internal benchmark of the honesty of the agents.”

Gemini app

In the Gemini app, Hydrogen is better able to follow and stick to your tone, format, and rules, while providing clear charts and visual breakdowns. This means “email drafts, study guides, rewrites, and documents come back ready to use the first time.” Additionally, Hydrogen “calls on the right tools (i.e. connected apps) and skills for the job more frequently.”

Advertising – scroll for more content

In the model selector, 4 Hydrogen is billed as providing “daily help.”

• Quick responses: 3.5 Flash-Lite
• Daily help: 4 Hydrogen
• Detailed search: 3.1 Pro (on AI Pro)
• Code and analyze:4Argons
• Most complex work: 4 Carbon

Free Gemini app users have access to Hydrogen via “(auto)model selection when it is the appropriate model for the prompt.”

Google Search

Google claims that Hydrogen significantly improves search (AI mode) by better understanding your intent:

…you can ask your question according to your natural flow of thought, even if your query has multiple parts, hard-to-understand typos, or very specific requests, and search will intelligently understand what you’re looking for.

Responses become more visual and will feature a wider range of relevant links. Hydrogen is the default model for AI mode and is used for “your toughest questions” in AI previews.

Antigravity

In Google Antigravity, 4 Hydrogen has an “Extra” reflection level alongside Low, Medium and High. Then there’s Orchestrator: “Carbon can transfer routine steps to hydrogen. Same quality carbon level, fewer tokens used.”

To help automate this multi-model collaboration, we are introducing Orchestrator into Google Antigravity. Rather than using a single template from start to finish, Orchestrator dynamically switches between templates like Carbon and Hydrogen within the same session.

For workflows that require parallel agents, Carbon can act as a coordinator while Hydgen can enable worker and review agents to create and test changes simultaneously in isolated workspaces, helping teams execute complex projects faster without increasing token costs.

Hydrogen is available in Antigravity for paid Google AI Pro and Ultra subscribers.

For developers, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens.

Update…