Google's new Gemini 4, with benchmark scores surpassing Argon, is free to use on the search engine. / CheapToken News

Upon waking up, Google has released the Gemini 4 model again. On November 12, Google officially launched Gemini 4 Hydrogen and simultaneously introduced the multi-model intelligent scheduling system Orchestrator. This is a large model at the Flash level (also the successor to Gemini 3.8 Flash), natively multimodal, with an input, thinking, and output window of 1 million tokens. However, according to Gemini's new naming convention, Google has chosen Hydrogen (hydrogen) as the level name. In the Gemini 4 product lineup, Carbon and Argon focus on deep reasoning and complex tasks, while Hydrogen serves as the "fast execution engine." Hydrogen emphasizes efficient operation and quick response, making it very suitable for daily coding, agentic workflows, and multi-step data processing. By using Hydrogen, Geminisland allows anyone to generate custom structures in real-time on a shared pixel island with simple instructions. Players can freely explore and walk around in first-person view. In multiple benchmark tests, Hydrogen is competitive in knowledge work, automation, and code fixing tasks: It is benchmarked against Anthropic's Claude Sonnet 5.5 and OpenAI's GPT-6.1 Terra. In practical tests involving complex graph computations, SQL queries, and high-concurrency enterprise data workflows, Hydrogen's performance speed has improved by 2-2.8 times compared to the previous generation; meanwhile, when it comes to cross-modal understanding of long videos and large codebases (with millions of context tokens), the first token latency (TTFT) remains at a low level. On the Artificial Analysis Intelligence Index, Hydrogen, in its default high mode, scored 51 points, and in its highest xhigh mode, it scored 54 points, which even surpassed Google's larger parameter Argon (50 points). It is evident that the Hydrogen model has added an xhigh level of reasoning strength. Google recommends starting with the high mode for main tasks, and using xhigh only when encountering difficult multi-step problems, as the latter may increase latency, token consumption, and overthink on simple tasks. By leveraging the Hydrogen model, Google significantly expanded the Gemini 4 series, making it accessible to everyone, including all levels of Google AI Studio, Gemini API, Gemini for Business, Gemini applications, Google Search, and the paid Google AI Pro and Ultra subscribers in Google Antigravity. However, the availability through different entry points varies. In the Gemini application, free users receive Hydrogen responses through the Auto model selection, while Google AI Plus, Pro, and Ultra users can manually select it. Developers can use the free tier of AI Studio and Gemini API, but are subject to rate limits. The API price for Gemini 4 Hydrogen is $1.5 per million input tokens, $0.075 for cache hits, and $7.5 per output token. Google stated that although developers need to use the advanced models at the Carbon and Argon levels for in-depth architectural decisions in actual software development, to design systems or resolve subtle bugs, they still require a fast and standardized execution model for coding, running tests, and verifying changes. The design of Hydrogen is intended to serve both as an independent model for daily work and as an efficient execution engine within a multi-model workflow when combined with Carbon and Argon.

To help achieve automation in multi-model collaboration, Google introduced Orchestrator in its programming tool Google Antigravity. This mechanism allows dynamic switching between models such as Carbon and Hydrogen within the same session. In the Terminal-Bench 4.0 test, Orchestrator achieved task success rates close to Carbon, reducing the average quota consumption by 24% and improving completion speed by 19%. For workflows requiring parallel agents, Carbon can act as the orchestrator, while Hydrogen can support both the agent and the verification agent, enabling simultaneous building and testing of changes in isolated workspaces. This helps teams deliver complex projects faster without increasing token costs.

The Orchestrator feature has been first rolled out to Ultra users.

In addition to multi-model workflows, Hydrogen represents a significant independent upgrade over Gemini 3.8 Flash. It has achieved notable improvements in speed and intelligence: under high default settings, task completion speed increased by 2.5 times, Terminal-Bench 4.1 scores rose more than threefold (from 18.2% to 57%), hallucinations decreased by more than five times, and it can seamlessly adapt when tool calls fail or search results are empty.

Efficiency improvements are also reflected in how the model uses tools. When exploring large codebases, Hydrogen can batch and execute file reading and search calls in parallel, reducing the waiting time caused by viewing files one by one. When necessary context is missing, it prefers to point out the information gap rather than guess.

Google demonstrated a more intuitive office scenario: by providing Hydrogen with a YouTube keynote video and a design template in Antigravity, it can directly generate a multi-page presentation in Google Slides, then automatically check and correct layout issues such as text overlap. From understanding the video, organizing content, to invoking applications and checking results, multimodal capabilities are integrated throughout the entire task.

What is it like to have a model that is faster than 3.8 Flash? It seems hard to describe. Google has now made Hydrogen the default model for the search engine AI mode, and uses it to support the underlying reasoning for AI Overviews on the search main interface, providing users worldwide with more practical and intuitive search results.

With the Hydrogen model, people can ask questions in the search box in a natural way, even if the query contains multiple parts, hard-to-understand spelling errors, or very specific requirements. The search can intelligently understand what you are looking for. In the search results, you will receive highly adaptive, visualized, and easy-to-browse feedback generated by AI.

Leveraging Hydrogen's advanced reasoning technology, Google also upgraded the query fan-out (Query Fan-Out, the mechanism where LLM breaks down complex questions and performs parallel retrieval and generation, then merges the results) in search, providing users with more extensive useful links than ever before, enabling them to find websites that were previously undiscovered.