aipoc-lab
/katTech/temAgents · Multimodal · Tooling/podGoogle · Meta/typNews

AI Today: Gemini 4 Hydrogen and AlphaGenome 2

Authorship and approval

[AI-generated content][Approved by a human]

Published and updated

Published: Updated:

State of knowledge as of 2026-11-13. Source material gathered between 2026-11-12 and 2026-11-13.

Google introduces Gemini 4 Hydrogen in developer tools and search, alongside the Decisions API for evaluating multiple questions at once. In genetic research, AlphaGenome 2 analyzes the impact of DNA changes in specific cell types. Meta, in turn, adds user-created spaces to Muse. Below we have outlined the key capabilities and limitations described in the announcements.

Highlights Today (TL;DR)

  • Gemini 4 Hydrogen: scored 51 points with the default high "thinking" level on the Artificial Analysis Intelligence Index.
  • Gemini 4 Hydrogen in Search: handles complex, multi-step queries, and responses can include tables, images, and links.
  • Gemini Decisions API: handles up to 1000 questions in a single pass across text, images, audio, video, and PDFs.
  • AlphaGenome 2: points to nearly twice as many gene-activity-altering variants as the previous model.
  • Muse: allows users to create custom spaces based on user descriptions.

Google Releases Gemini 4 Hydrogen with a Score of 51 Points

Google has released Gemini 4 Hydrogen for everyday coding and working with AI agents. Developers can use it via the Gemini API, Google AI Studio, and Android Studio; with the default high "thinking" level, it scored 51 points on the Artificial Analysis Intelligence Index, compared to 50 for Argon.

  • At the highest xhigh setting, Hydrogen scores 54 points on this index.
  • The context window spans 1 million tokens. The price is 1.50 USD per million input tokens and 7.50 USD per million output tokens.
  • The orchestrator switches between the Carbon and Hydrogen models. In Terminal-Bench 4.0, it achieved an effectiveness close to Carbon while consuming an average of 24% less rate limit and finishing tasks 19% faster.
  • Compared to Gemini 3.8 Flash, Hydrogen finishes tasks 2.5 times faster and more than triples the Terminal-Bench 4.1 score: 57% versus 18.2%.

In the Lab: We wrote about Gemini 4 Argon, a model for the hardest problems, in our post "AI Today: NVIDIA RTX Spark and Gemini 4 Argon". Hydrogen complements this offering as a model for faster execution of everyday tasks.

For AI Builders: Teams building programming tools can test Hydrogen on repository coding and debugging. Sources do not mention Polish language support, API limits, or EU availability.

Google Deploys Gemini 4 Hydrogen Globally in Search

Google is bringing Gemini 4 Hydrogen to Search to handle complex, multi-step queries. Users can ask in natural language about travel, shopping, and hobbies, receiving answers featuring tables, images, and links.

  • Gemini 4 Hydrogen recognizes intent and handles multi-part queries.
  • The model can run parallel queries and organize answers into tables and images.
  • Google highlights queries about specific travel, shopping, and hobby requirements.
  • The update is available globally to all Search users.

In the Lab: We wrote about Gemini 4 Argon in the Gemini app; now Google is introducing a Gemini 4 family model to Search as well.

For AI Builders: Assistant creators and product teams can check whether table and image formatting fits their scenarios. The announcement does not provide pricing, limits, or details on availability in Poland and the EU.

Google Gemini Decisions Evaluates 1000 Questions in a Single Pass

Google has released the Gemini Decisions API with the Decision 1 model: up to 1000 questions in a single pass across text, images, audio, video, and PDFs. Developers can use it for tool selection, rule checking, or content evaluation.

  • Questions: The API supports yes/no answers, option selection, and criterion-based scoring.
  • Input: A single format mix can contain up to 256k tokens, meaning units of text processed by the model.
  • Time and Price: An answer takes 150–250 ms; 1 million input tokens cost 0.10 dollars, and output tokens cost zero.
  • Confidence: Every answer receives a score from 0 to 1, which can help route ambiguous cases to a human.

In the Lab: On October 23 we wrote about OpenAI's Decisions API; Google now offers a similar type of API for fast, constrained choices.

For AI Builders: Teams building agent routing or compliance checks can test the API on their own data. Google does not mention Polish language support or EU availability in this announcement.

Google AlphaGenome 2 Nearly Doubles the Detection of Variants Affecting Expression

Google announced AlphaGenome 2, a model predicting how DNA changes impact gene activity in specific cell types; it indicates nearly twice as many activity-altering variants as the previous model. Genetics researchers can check the effects of changes across hundreds of cell types instead of relying on averaged tissue samples.

  • Over 107 million single-cell profiles from 68 tissues were used for training; data came from humans and mice.
  • The model analyzes one designated cell type or state along with the DNA sequence at once.
  • Researchers can select a cell from the catalog or provide data from their own single-cell experiment.
  • Based on a single measurement, e.g., gene activity, the model can predict other traits, e.g., chromatin accessibility — the structure in which a cell packages DNA.
  • The API is available for non-commercial research; commercial access remains in private preview.

In the Lab: The lab database features AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome from September 9, mapping the effects of 9 billion single DNA letter changes; AlphaGenome 2 takes the analysis down to specific cell types.

For AI Builders: Bioinformatics teams can test the API on their own single-cell data. The announcement does not specify pricing or terms for processing uploaded data.

Meta Allows Users to Create Custom Spaces in the Muse App

Meta is adding "spaces" to Muse: custom spaces created based on user descriptions. Users can develop ideas and routines within them and then return to them inside the app.

  • An idea description is enough for Muse to build a space. Afterward, its content, appearance, and behavior can be modified.
  • Spaces can be moved between the sidebar and main tabs, hidden, and scheduled for specific appearances.
  • Once connected apps are linked, spaces update in the background. Users can also add notes and make changes inside them.
  • Spaces are available wherever Muse runs. Meta is also exploring sharing them with others and is rolling out special model versions developed from Muse Wonder experiences.

In the Lab: We previously wrote about voice conversations in Muse in the article Give Your Muse a Voice. Meta is expanding the app with diverse interaction modes: voice and custom spaces.

For AI Builders: AI app designers can check which integrations are supported and in which markets Muse operates. The announcement does not provide a list of connected apps, pricing, or language availability.

Frequently Asked Questions (FAQ)

Where can Gemini 4 Hydrogen be used?

Developers can use the model via the Gemini API, Google AI Studio, and Android Studio. Google released it for everyday coding and working with AI agents; it scored 51 points on the Artificial Analysis Intelligence Index at the default high thinking level.

What data does the Gemini Decisions API accept?

The Gemini Decisions API accepts text, images, audio, video, and PDFs, up to 256k tokens in a single format mix. It can answer yes or no, select options, or evaluate content according to criteria.

How do users create custom spaces in Muse?

A user describes an idea, and Muse creates a space whose content, appearance, and behavior can subsequently be changed. Spaces can be moved between the sidebar and main tabs, hidden, and scheduled for appearance.

Sources

--- Text prepared automatically by the aipoc-lab (AI) laboratory based on publicly available sources listed above. Own summaries, not copies. Pre-human-editing version.

Provenance manifest of this post (JSON) — which signals it was built from, what decisions the loops made, and when a human approved it.

Related posts

/katTech/temMultimodal · Open Source · Tooling/podDeepSeek · Google/typNews

AI Today: DeepSeek V4.1-Flash and Kimi K3.1 Pro

Discover the latest updates on DeepSeek V4.1-Flash, Kimi K3.1 Pro, Muse voice chats, Apple CarPlay integration, and Gemini Omni 1.2 Flash.

/katTech/temAgents · Multimodal · Robotics/podGoogle · OpenAI/typNews

AI Today: Gemini 4 Carbon and Gemini Robotics 2.5

Google releases Gemini 4 Carbon and Gemini Robotics 2.5, while OpenAI launches business profiles and agents in ChatGPT.

Analysis
/katTech/temAgents · LLM · Tooling/podGoogle · OpenAI/typAnalysis

API Architecture and Ecosystem in Agentic Systems

An analysis of API architectures, standardization protocols, and performance optimization in modern agentic AI systems.

/katTech/temAgents · LLM · Tooling/podGoogle · OpenAI/typNews

AI Today: Qodo, OpenAI Pando, and Gemini 4

Stay updated on the latest developments in developer tools, infrastructure, and new artificial intelligence models from Qodo, OpenAI, and Google.