Gemini 4 Hydrogen now available on AI Gateway

Google’s Gemini 4 Hydrogen model is now accessible via Vercel’s AI Gateway, replacing the previous Gemini 3.8 Flash. The upgrade focuses on enhanced agentic coding, knowledge work, and tool use while maintaining the same latency class. It supports longer engineering tasks and produces more accurate documents, spreadsheets, and presentations with fewer agent steps. The model adheres more closely to system instructions, formatting requirements, and task boundaries, improving tool orchestration in extended conversations.

Gemini 4 Hydrogen introduces stronger visual reasoning and technical problem-solving capabilities, supporting multimodal workflows. It also reduces unsupported claims in search-backed answers and minimizes invented tool or test results in coding reports. The model is designed to follow grounding more effectively, ensuring higher reliability in outputs. Users can access it through supported coding agents like Claude Code, Codex, and Cursor.

To use Gemini 4 Hydrogen, users must install the latest Vercel CLI and run the setup process, then select the model in their agent. The model is also available for testing in the AI Gateway model playground. Vercel provides documentation and guides to assist with integration and usage. The availability follows Google’s announcement of the model’s general availability.

AI Gateway serves as a unified API for model access, offering usage tracking, cost management, and configuration options like retries, failover, and performance optimizations. It includes features such as Zero Data Retention support, custom reporting, API key budgets, and routing rules. Pricing aligns with provider rates, with no platform fee on inference, including Bring Your Own Key requests. Users can view all available language models on the platform.