AI

Google updates Gemini 4 lineup and unveils new AI models

Google announced updates to the Gemini 4 lineup, Gemini Omni 1.2 Flash video generator, and Gemini Robotics 2.5 for robotic control.

·3 min read
Google updates Gemini 4 lineup and unveils new AI models

Over recent weeks, Google has significantly updated its artificial intelligence ecosystem. The updates cover the Gemini 4 model family, video generation tools, and robotics. These new solutions aim to boost performance, reduce costs, and expand agentic programming capabilities.

  • Google updated its Gemini 4 lineup with the introduction of Hydrogen and Carbon models.
  • Released Gemini Omni 1.2 Flash for video generation with longer durations and lower pricing.
  • Announced Gemini Robotics 2.5, a new vision-language-action model for robotic control.
  • Introduced new orchestrators and parallel agent workflows in Antigravity.

Gemini 4 Hydrogen and Carbon: balancing speed and complexity

On November 12, Google introduced the Gemini Hydrogen model, designed for execution speed and resource efficiency. API pricing is set at $1.50 per 1 million input tokens and $7.50 for output tokens. Google also created an orchestrator for Antigravity that automatically switches between Carbon and Hydrogen within a single session.

On November 10, Google made the Gemini 4 Carbon model widely available for complex agentic workflows and scientific research. Initial API pricing starts at $2 per 1 million input and $10 for output tokens, with planned increases after November 30, 2026. Additionally, Google launched preview testing of /swarm in Antigravity for parallel agent execution.

Read alsoAntigravity /teamwork: Google multiagent AI systems for science

Gemini Omni 1.2 Flash: advanced video generation features

On November 11, Google updated the Gemini Omni 1.2 Flash video generation model. It now supports video generation up to 15 seconds instead of the previous 10, while generation costs decreased by 30%. 720p video generation now costs $0.07 per second.

The update improves the consistency of character appearances, objects, and clothing details across frames. It also introduces multi-step video editing via dialogue, allowing users to alter backgrounds, lighting, and weather sequentially without regenerating the entire video.

Gemini Robotics 2.5 and robotic integration

On November 10, Google DeepMind introduced Gemini Robotics 2.5, a new vision-language-action model for robotic control. It provides enhanced stability in physical action sequences and improved resilience against adversarial jailbreak scenarios.

Tested across more than 200 tasks, success rates increased by 30–100%, and speed improved by 20–30%. A practical demonstration took place on a Mercedes-Benz production line, where the Apollo 2 humanoid robot by Apptronik sorted automotive parts with a 96% success rate.

What this means for developers and the industry

Google's latest releases mark a shift towards specialized AI ecosystems tailored for practical implementation. The combination of Hydrogen and Carbon models helps optimize infrastructure costs for complex applications, while advancements in video generation and robotics drive automation across industries.

Source: Все публикации подряд на Хабре

Related articles