AI Daily | October 28, 2026

Model Releases & Updates

Google DeepMind Announces Genie 4 Interactive World Model GGoogle DeepMind

• Bottom line：Google DeepMind unveiled Genie 4, a real-time interactive world model combining physical simulation, reasoning, and native audio generation.
• Core Highlights：Delivers real-time interactive simulations with enhanced realism and dynamic environment exploration.
• Audio Integration：Natively generates ambient sound effects and interactive character speech within simulated virtual worlds.
• Source：Google DeepMind Blog

Liquid AI Releases LFM3 Open-Weight Multimodal Models LLiquid AI

• Bottom line：Liquid AI released the LFM3 series, featuring six open-weight 2-billion parameter multimodal models built on a tri-hybrid architecture.
• Model Lineup：Includes six open-weight 2B models designed for diverse multi-domain tasks.
• Unified Architecture：Covers text, vision, audio, omni, and decision-making capabilities using a shared tri-hybrid backbone.
• Source：Hugging Face Blog

Product Releases & Updates

Gemini Robotics Embodied Reasoning 2 Available on Google Cloud GGoogle Cloud

• Bottom line：Google Cloud launched Gemini Robotics Embodied Reasoning 2 in public preview to serve as an advanced cognitive brain for robotic systems.
• Capabilities：Combines spatial and temporal video understanding with long-horizon planning and safety guardrails.
• Enterprise Deployment：Allows organizations to orchestrate multi-minute robotic tasks and cross-robot coordination in production environments.
• Source：Google Cloud Blog

Replit Opens Cloud Agents to Pro and Enterprise Customers RReplit

• Bottom line：Replit expanded access to Cloud Agents, enabling non-technical staff to submit fully tested pull requests using natural language.
• Target Audience：Empowers design, customer support, and other non-engineering team members to contribute directly to company codebases.
• Workflow Automation：Automates testing and validation before submitting pull requests for organizational review.
• Source：Replit Blog

ElevenLabs Audio Models Integrated into Vercel AI Gateway VVercel / ElevenLabs

• Bottom line：Vercel partnered with ElevenLabs to bring 14 audio models to AI Gateway alongside an introductory pricing discount.
• Model Support：Supports speech-to-text, streaming text-to-speech, real-time voice agents, music, and sound effect generation.
• Launch Promotion：Offers a 50% discount on all ElevenLabs models through November 9.
• Source：Vercel Blog

Industry News

Google TPUs to Deploy on Oracle Cloud Infrastructure Starting 2027 GGoogle Cloud & Oracle

• Bottom line：Google Cloud and Oracle announced a partnership to bring Google Tensor Processing Units to Oracle Cloud Infrastructure beginning in 2027.
• Strategic Partnership：Makes OCI the first hyperscaler outside of Google to offer Google TPUs for enterprise AI workloads.
• Deployment Timeline：Capacity rollout is scheduled to begin in calendar year 2027 and scale progressively.
• Source：Google Cloud Blog

Anthropic Expands GSA Partnership to Offer Claude at 40% Discount to Government AAnthropic

• Bottom line：Anthropic expanded its agreement with the U.S. GSA to provide Claude Enterprise, Government, and API at a 40% discount across all government levels.
• Agreement Scope：Utilizes the new OneGov protocol to supply government entities with discounted AI tooling.
• Eligible Bodies：Covers federal, state, local, tribal, and territorial government organizations across the United States.
• Source：Anthropic Announcement

OpenAI Partners with Check Point to Accelerate Vulnerability Research OOpenAI / Check Point

• Bottom line：OpenAI announced a partnership utilizing OpenAI Daybreak to help Check Point analysts accelerate vulnerability research and security defenses.
• Tool Integration：Employs OpenAI Daybreak to speed up the analysis behind new security protections.
• Operational Impact：Enables security teams to turn vulnerability intelligence into customer defenses faster.
• Source：OpenAI Announcement

Research Papers

Ai2 Releases NoveltyBench to Evaluate LLM Research Novelty Judgments AAllen Institute for AI (Ai2)

• Bottom line：Ai2 released research and a benchmark evaluating the reliability of large language models when judging the novelty of scientific research ideas.
• Research Focus：Investigates whether LLMs can be trusted to accurately evaluate whether a scientific research idea is novel.
• Open Resources：Includes benchmark data, code, and automated pipelines to facilitate transparent peer audits.
• Source：Ai2 Research Blog

UC Berkeley Open-Sources LEANN Vector Index for Massive RAG Compression UUC Berkeley Sky Computing Lab

• Bottom line：UC Berkeley open-sourced LEANN, a vector index that reduces a 201 GB Wikipedia index down to 6 GB by recalculating embeddings at query time.
• Compression Ratio：Successfully compresses a 60-million chunk RAG index from 201 GB to just 6 GB.
• Mechanism：Recomputes embeddings during query execution rather than storing pre-calculated vector representations.
• Source：MarkTechPost Coverage

Other Highlights

JetBrains IDEs Terminal Now Powered by libghostty JJetBrains

• Bottom line：JetBrains updated the built-in terminal in JetBrains IDEs to run on libghostty, supporting modern terminal protocols and demanding AI tools.
• Library Replacement：Replaces JediTerm with libghostty, the terminal emulation library powering Ghostty, starting with version 2026.3.
• Protocol Support：Enables native support for advanced TUI applications and AI coding assistants like Claude Code and Codex.
• Source：JetBrains Blog