Google lança o Gemini 4 Hydrogen: modelo rápido para fluxos agênticos e para o Search, com ganhos de segurança
Matéria consolidada. Reúne 2 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 2 fontes trouxeram em 14/11/2026 00:02; os links antigos trazem para cá.
On November 12, 2026, Google (Google DeepMind) unveiled Gemini 4 Hydrogen, a model built for fast, cost-efficient execution of everyday agentic workflows (securityonline.info). It is described as running 2.5 times faster than its predecessor Gemini 3.8 Flash and outperforming the more powerful Gemini 4 Argon on certain intelligence benchmarks (techcrunch.com, securityonline.info). Google is using it to improve Google Search and AI Mode on complicated multi-part questions and typos, with better user intent recognition, enhanced fan-out techniques to discover relevant websites, and richer visual responses such as images and comparison tables (techcrunch.com). It also powers Workspace tasks, with better query parsing, tool calling and cross-lingual understanding (securityonline.info). A global rollout was already underway as of the first report (techcrunch.com). For defenders, the most relevant claims are the reported reductions in agent overreach and stronger defenses against prompt injection, which come from Google's own figures and have not been independently verified in the sources.
Detalhes técnicos
- Context and pricing: one-million-token bidirectional context window, with flat pricing across the full context length (securityonline.info).
- Reasoning control: four thinking or reasoning levels to balance speed and thoroughness (techcrunch.com, securityonline.info).
- Availability: Google AI Studio, Gemini API, Google Antigravity, Workspace and Search (techcrunch.com, securityonline.info).
- Safety claims: unauthorized agent overreach reduced by 81%, stronger defenses against prompt injection attacks, and high honesty and legibility maintained in audit trials (securityonline.info). These are vendor-reported figures; the sources give no methodology or baseline.
Linha do tempo
-
Google Launches Gemini 4 Hydrogen Model for High-Speed AI Workflows with Advanced Safety Features
ler a análise desta fonte ocultar
On November 12, 2026, Google DeepMind unveiled Gemini 4 Hydrogen, a model designed for rapid, cost-efficient execution of everyday agentic workflows. It offers a one-million-token bidirectional context window, supports four reasoning tiers, and features flat pricing across its extensive context length. Hydrogen surpasses the performance of its predecessor Gemini 3.8 Flash and even outperforms the more advanced Gemini 4 Argon model in certain benchmarks. It powers Google Search AI Mode and Workspace tasks, enhancing query parsing, tool calling, and cross-lingual understanding. Significant safety improvements include reducing unauthorized agent overreach by 81%, stronger defenses against prompt injection attacks, and maintaining high honesty and legibility in audit trials. Hydrogen is integrated within Google AI Studio, Gemini API, Google Antigravity, Workspace, and Search.
-
Google deploys Gemini 4 Hydrogen to enhance Search with faster, nuanced understanding of complex queries
ler a análise desta fonte ocultar
Google has released its Gemini 4 Hydrogen AI model to improve Google Search's handling of complicated, multi-part questions and typos, promising faster and more precise responses. Hydrogen operates 2.5 times faster than its predecessor Gemini 3.8 Flash and outperforms the more powerful Gemini 4 Argon on certain intelligence benchmarks. Key improvements include better user intent recognition, enhanced fan-out techniques to discover relevant websites, and richer visual responses such as images and comparison tables. Hydrogen supports four thinking levels for balancing speed and thoroughness, and it integrates into multiple Google AI platforms including AI Studio, Gemini API, and Antigravity, with a global rollout already underway.