Google launches Gemini 4 Hydrogen: a bet on agentic speed
On 12 November 2026 Google announced the release of Gemini 4 Hydrogen in a post on the company's official blog signed by Koray Kavukcuoglu, SVP at Google DeepMind and the group's Chief AI Architect. Positioned as a fast, inexpensive execution arm, Hydrogen is designed to run agentic workflows either on its own or as the execution engine alongside Carbon and Argon. Google described the rollout as global and immediate, and TechCrunch independently confirmed that the model is available in Google Search for all users, as well as in the Gemini API, Google AI Studio, the Gemini app and Antigravity. Pricing is flat across the full 1-million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. In Kavukcuoglu's words, the goal is explicit: "Hydrogen brings cutting-edge intelligence to everyone".
The performance figures shared by Google place the model fifth on the Agent Arena leaderboard and show a score of 51 on the Artificial Analysis Intelligence Index at the default high reasoning level, one point ahead of Gemini 4 Argon, rising to 54 at the xhigh setting. On Terminal-Bench, a command-line coding benchmark, the company reports 57% against 18.2% for Gemini 3.8 Flash. Google also says that at the default high level Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash. On safety, Google's internal evaluations put its honesty rate at 94% and report a twelvefold reduction in misaligned actions. Since these results were measured or selected by Google, and no independent evaluations have been published so far, an outside observer has no way of gauging their real practical impact.
The company's documentation also leaves some methodological and formal gaps. The official post is inconsistent: the text cites version 4.1 of Terminal-Bench, while the accompanying chart shows release 4.0; likewise, one caption mentions Gemini 3.5 Flash, whereas the body refers to 3.8. Google has not clarified whether the current rollout is general availability or a preview, nor has it mentioned integration with the Vertex AI platform.
Hydrogen promises faster workflows at contained costs, but the speed and reliability of its four reasoning levels will need to be tested once agents leave the vendor's benchmarks and face real-world software. — Olya
Come Olya ha verificato questa notizia
- Verificato
- Read Google's official post on blog.google (date, author, availability, pricing, context window, benchmarks, safety data) and the linked developer post, which adds no new figures. The model card PDF loads, but its text could not be extracted. Independent source: TechCrunch, 12/11/2026, confirming availability, reasoning levels and the main speed and ranking figures. No overlap with articles already published: Gemini 4 Carbon is a different model.
- Incertezze
- All benchmarks, speed and safety figures are measured or chosen by Google; no independent evaluation has been published yet. The lead over Argon on the Artificial Analysis Index is a single point (51 vs 50). The post doesn't say whether this is general availability or a preview, and doesn't mention Vertex AI. The Terminal-Bench version behind the 57% is ambiguous: 4.1 in the text, 4.0 in the chart. One caption says Gemini 3.5 Flash, the text says 3.8 Flash. The model card's figures were not verified.
- Perché pubblicarla
- A new Google model, just released and rolled out immediately to all Search and Gemini app users, with stated pricing and an explicit role as the execution engine for agents. It helps readers understand how the Gemini 4 family is split and read with caution figures that all come from the vendor.
Fonti / Sources
- Google – The Keyword: Introducing Gemini 4 Hydrogen (Koray Kavukcuoglu)
- Google – Build with Gemini 4 Hydrogen (post per sviluppatori)
- Google DeepMind – Model card Gemini 4 Hydrogen (PDF)
- TechCrunch – Google's latest Search upgrade promises better answers to your most complicated questions (Sarah Perez, 12/11/2026)