Google replaces Gemini 3.8 Flash with faster Gemini 4 Hydrogen

A new standard for speed

Google has officially released Gemini 4 Hydrogen, a new model within its Gemini 4 family designed for efficiency. This model is specifically built to handle everyday work, coding, and agentic tasks that do not require deep reasoning. It serves as a faster and more reliable alternative to the previous Gemini 3.8 Flash model.

The rollout is already underway across several major platforms. Users worldwide will see Hydrogen powering responses in Google Search. It is also being integrated into the Gemini app, Google Sheets, and various developer tools like Google AI Studio and Android Studio.

According to the company, Hydrogen is significantly more performant than its predecessor. Google reports that the new model finishes tasks 2.5 times faster than Gemini 3.8 Flash. This speed improvement is a core part of the update for users who need quick results without sacrificing accuracy.

Improvements in Search and Sheets

Google Search users will notice several functional changes with the integration of Hydrogen. The model is designed to handle multi-part questions and typos more effectively than previous versions. It also aims to provide more visual answers and a broader range of website links in search results.

In the Gemini app, the model is designed to follow formatting, tone, and other specific instructions more closely. It is also built to maintain these instructions throughout longer conversations. This makes it more reliable for complex prompts that require a specific style or structure over several turns.

The update also impacts productivity tools like Google Sheets. For users with Google AI Pro, Ultra, or eligible Workspace accounts, Hydrogen can turn financial data into interactive forecasting models. Google states that this specific process is now more than 30 percent faster than it was with previous models.

Flexible reasoning and pricing

One of the most notable technical features of Hydrogen is its ability to scale based on user needs. The model offers four distinct thinking levels: low, medium, high, and a new xhigh level. This allows users to choose the level of depth required for a specific prompt, which helps manage both time and costs.

For developers using the Gemini API, the model comes with a 1 million token context window. A token is a basic unit of text that the AI processes to understand context. The pricing for the API is set at $1.50 per million input tokens and $7.50 per million output tokens.

Benchmarks show the model's performance across these levels. On the Artificial Analysis Intelligence Index, Hydrogen scored 51 at its default thinking level. It scored 54 at its highest setting, which is just one point ahead of the Gemini 4 Argon model, which scored 50.

Safety and agentic behavior

Google has released specific safety figures to highlight how Hydrogen handles autonomous tasks. The company claims that Hydrogen reduces agentic overreach by up to 81 percent when compared to Gemini 3.8 Flash. Overreach occurs when an AI performs actions beyond the intended scope of a user's request.

The model also shows significant improvements in following rules and honesty. It is reported to cut misaligned shortcuts by 12 times and is up to twice as unlikely to bypass permission restrictions. These metrics are intended to make the model safer for use in automated workflows and agentic tasks.

In internal agent tests, Hydrogen reached a 96.7 percent honesty rate. These improvements are part of Google's effort to make the model more reliable for complex, multi-step tasks. It is also designed to pick the right tool more often when connected to apps like Gmail, Google Drive, Dropbox, and Airtable.

Orchestration and integration

For more complex workflows, Google is introducing a new Orchestrator feature in Google Antigravity. This feature allows the system to switch between different models depending on the task at hand. It uses Gemini 4 Carbon for the planning phase and Gemini 4 Hydrogen for the execution phase.

This dual-model approach is designed to optimize performance and resource usage. Google says this specific orchestration cuts quota usage by 24 percent. It also speeds up the completion of tasks by nearly 20 percent by using the right model for the right job.

While Hydrogen is now the primary model for many Google products, the transition is ongoing. Google has not yet announced a specific date for when Gemini 3.8 Flash will be removed from the API. For now, the new model is the standard for most users interacting with Google's AI ecosystem.