zhineng.dev
English
Get started
← All models

Gemini 4 Hydrogen

Published Publisher Googlegemini-4-hydrogen

Overview

Gemini 4 Hydrogen is Google’s multimodal model designed for fast, efficient everyday coding and agentic workflows. It can operate independently or serve as an execution model alongside Gemini 4 Carbon and Argon. Google says it is adapted from Gemini 4 Carbon. (blog.google)

Capabilities and uses

  • Accepts text, images, audio and video. The model card specifies a combined input-output context window of up to 1 million tokens, with up to 64,000 synchronous output tokens by default or up to 1 million through background long decode. (deepmind.google)
  • Offers four thinking levels: low, medium, high (default) and xhigh, allowing users to trade off reasoning depth, latency and cost. (blog.google)
  • Intended uses include repository coding and debugging, multi-step agentic execution, multimodal research and knowledge work, web development, writing and prototyping. It can also work as a sub-agent in multi-model workflows. (blog.google)
  • Google reports a 57.0% score on Terminal-Bench 4.1 and 71.6% on Agents’ Last Exam v1.1; these are Google-reported evaluations, not guarantees of real-world performance. (deepmind.google)

Availability and pricing

Google announced availability through Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and Google Antigravity for paid Google AI Pro and Ultra subscribers. Availability and limits may vary by product and plan. (blog.google)

The launch announcement lists pricing in US dollars at $1.50 per million input tokens and $7.50 per million output tokens, with the same rate across the 1-million-token context window. Pricing can differ by platform and service configuration. (blog.google)

Limitations

The model may hallucinate and remains subject to jailbreak risks. Google also notes occasional slowness or timeouts and potentially higher token usage, especially at higher thinking levels. Its knowledge cutoff is March 2026, although knowledge in some domains may be less current. (deepmind.google)

Release and official sources