Google launches Gemini 4 Hydrogen, a faster model for coding and agent tasks, presented as an update to Gemini 3.8 Flash

Google launches Gemini 4 Hydrogen, a faster model for coding and agent tasks, presented as an update to Gemini 3.8 Flash

On November 12, Google announced the global rollout of Gemini 4 Hydrogen across its products for developers, businesses, and the general public. According to the company, it is a faster model from the Gemini family, designed for coding and other agent-based tasks, meaning programs capable of chaining multiple actions to achieve a goal.

Gemini, sometimes marketed as Google Gemini and initially known as Bard, is a generative AI-based chatbot and virtual assistant developed by Google. It is built on the large language model (LLM) family of the same name, which previously relied on LaMDA and PaLM 2. The chatbot can process and generate text, computer code, images, audio files, and videos simultaneously.

A fast model, accessible to all

Google presents Hydrogen as intended for fast execution in various agent-based workflows, while its flagship models, Gemini 4 Carbon and Gemini 4 Argon, are described as more suited to deep reasoning on complex problems. The model is available across all tiers of Google AI Studio, the Gemini API, Gemini for Business, the Gemini app, and Google Search, as well as for paid Google AI Pro and Ultra subscribers in Google Antigravity.

Google also introduced Orchestrator, a Google Antigravity feature that allows switching between multiple models, such as Carbon and Hydrogen, during a single session. On the Terminal-Bench 4.0 test, the company states that Orchestrator achieves success rates close to those of Carbon, while consuming on average 24% less quota and executing tasks 19% faster.

Results announced by Google

On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high reasoning level, compared to 50 for Argon, and reaches 54 at its “xhigh” setting, the highest. Google also claims the model performs tasks 2.5 times faster with its default setting and scores 57% on the Terminal-Bench 4.1 test, compared to 18.2% for Gemini 3.8 Flash. On Agent Arena, the real-time benchmark from arena.ai, it ranks fifth, and it achieves 58.8% on HLE-Diamond, where it leads its category.

Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens, within a one-million-token context window. The token, a unit of text processed by a model, serves as the basis for billing. According to Google, the model’s four reasoning levels (low, medium, default high, and xhigh) allow users to ‘pay only for the depth of reasoning needed for each task.’

However, excerpts from the public benchmarking methodology and API pricing pages gathered as part of this analysis contained no usable information. This prevented any independent verification of the methodology, availability, pricing, and API access levels for Hydrogen.

For individuals, Google highlights Gemini 4 Hydrogen’s ability to better follow instructions and select appropriate tools within the Gemini app. In Google Search, the model is intended to allow users to ask questions in natural language, including with typos or multiple parts, and receive more visual answers.

S’abonner
Notification pour
guest

0 Commentaires
Le plus ancien
Le plus récent Le plus populaire
0
Nous aimerions avoir votre avis, veuillez laisser un commentaire.x