THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

AI / NEWS BRIEF

Google launches Gemini 4 Hydrogen for faster agent workflows

Hydrogen brings the Gemini 4 family to Google's free and paid products with a flat API price across a one-million-token context window. Its benchmark advantages remain provider-reported.

By AIQuantumScore · · Event 2026-11-12

A faster member of the Gemini 4 family

Google has released Gemini 4 Hydrogen, a model adapted from Gemini 4 Carbon and optimized for rapid execution across coding and agentic workflows. It supports text, images, audio and video, with a combined input-and-output context limit of about one million tokens. Google says Hydrogen can operate at four thinking levels and positions it below the Carbon and Argon models for tasks that prioritize speed and efficiency over the longest reasoning runs.

Broad access and flat pricing

Hydrogen is available through Google AI Studio and the Gemini API, Gemini for business, the Gemini app and Google Search, with access also offered to paid Google AI Pro and Ultra subscribers in Antigravity. Google prices the API at $1.50 per million input tokens and $7.50 per million output tokens across the full context window. The developer release also describes repository coding, debugging, multimodal work and use alongside Carbon or Argon in multi-model systems.

The practical test is useful work per dollar

OUR ANALYSIS

Google reports that Hydrogen exceeds Argon on one third-party intelligence index at its default high thinking setting and improves further at the highest setting. Those figures come from Google's launch material and do not establish performance across every workload. The consequential part of this release is the combination of broad distribution, configurable reasoning and lower API pricing. Independent testing is still needed to determine latency, reliability and total cost on real agent tasks; AIQuantumScore will not assign a REVAL score without running the released model through the benchmark.

More to follow