$ cat wiki/models/gemini-4-hydrogen.md
Gemini 4 Hydrogen
Spec
| Attribute | Value |
|---|---|
| Developer | Google DeepMind |
| Released | 2026-11-12 |
| Announced | 2026-11-12 |
| Context window | 1 million tokens |
| Pricing | $1.50 per 1M input tokens; $7.50 per 1M output tokens |
| License | unknown |
| Availability | Gemini API, Google AI Studio, Gemini for business, Gemini app, Google Search; eligible paid subscriptions in Google Antigravity |
| Google states flat pricing across the context window and up to 1 million output tokens. The captured announcement does not establish a weight license; product eligibility and rollout differ by surface. (source) |
Release Date
Announced 2026-11-12; the article was updated 2026-11-13. Google describes a global rollout starting on the announcement date. (source)
Benchmarks
Google reports 57% on Terminal-Bench 4.1, versus 18.2% for Gemini 3.8 Flash, and task completion 2.5× faster at Hydrogen's default high setting. Separately, Antigravity Orchestrator approaches Carbon's task success on Terminal-Bench 4.0, with 24% less quota and 19% faster completion on average. These are different benchmark versions and vendor comparisons; quota is not a dollar cost. (source)
Use Cases
Everyday coding, tool execution and multimodal knowledge work. Four thinking levels are available: low, medium, high (default) and xhigh. Orchestrator can switch models within a session, while Carbon can coordinate Hydrogen workers and verifiers in isolated workspaces. These are documented capabilities, not a reproduced deployment evaluation. (source)
Compared To
Google positions Gemini 4 Carbon and Gemini 4 Argon for deeper architectural judgment and Hydrogen for faster execution. Its article's live Arena ranking should not be substituted into an older leaderboard capture; the publisher date of the local weekly snapshot predates Hydrogen's launch. (source) (dated snapshot)