Back to feed

Google DeepMind presents Gemini 4 Hydrogen model and Orchestrator mode

Google DeepMind has introduced the new Gemini 4 Hydrogen model and the Orchestrator mode in the Antigravity platform. The new solution combines high speed with cost-efficiency, offering an alternative to more resource-intensive versions of the lineup.

Performance and Benchmarks

The model demonstrates significant progress compared to its predecessors. In the Terminal-Bench 4.1 test, it scored 57.0%, while Gemini 3.8 Flash showed only 18.2%. At the same time, task generation and completion have been accelerated by 2.5 times.

According to the Artificial Analysis Intelligence Index, the basic mode received a score of 51, and at maximum reasoning depth — 54 points. This result surpasses the performance of the Argon model, which scores 50 points.

Tariffs and Positioning

The cost of using Gemini 4 Hydrogen is $1.50 per input and $7.50 per output. This makes its price equal to 37.5% of the cost of Carbon and Argon models, which will be priced at $4.00 and $20.00 respectively starting December 1.

Hydrogen is aimed at fast execution of everyday tasks, code writing, and processing subtasks within agentic systems. Technically, it is based on the Gemini 4 Carbon architecture but is not a direct successor to Flash.

Orchestrator Mode Features

The Orchestrator mode, available only in the Ultra plan, dynamically distributes subtasks between Carbon and Hydrogen models. Google claims a 24% reduction in quota usage and a 19% speedup without compromising effectiveness.

The model supports four levels of thinking (thinking), including the xhigh parameter. It is already deployed in Gemini API, Google AI Studio, Vertex AI, Android Studio, Google Search, and the Gemini app.

5.6K views

More from this channel Metavershche and AIshche

Similar in this category Technology