🤖 AI HOT Daily · Jun 24, 2026

FastWan-QAD generates 5s video in 1.8s on single GPU, JD.com's JoyAI full-stack interactive model, Confucius4-TTS 14-language voice cloning, Mistral OCR 4, ByteDance Seed2.1, IBM open-sources CUGA agent framework, Five Eyes warns of AI cyber threats.

1. 1. FastWan-QAD: 5s Video in 1.8s on a Single RTX 5090

   Sky Computing Lab releases a video generation model using quantization-aware distillation. End-to-end 5s 480P video in 1.8s on a single RTX 5090. Open-sourced.

2. 2. JD.com Open-Sources JoyAI-VL-Interaction: See and Talk

   World's first full-stack interactive model that continuously observes video streams, proactively identifies key events, and responds in real-time. Beats Doubao 77.6% and Gemini 87.9% in blind tests.

3. 3. Confucius4-TTS: 14-Language Accent-Free Voice Cloning

   NetEase Youdao releases 14-language accent-free voice cloning model. Zero-shot cloning from just 3s audio, 85%+ similarity, 97% task accuracy. Fully open-sourced under Apache.

4. 4. Doubao Audio Generation Model 1.0 Released

   Volcengine launches, supporting text and reference audio generation. Single prompt can orchestrate multi-character dialogue, emotion, BGM, and ambiance. API beta open.

5. 5. Mistral OCR 4: Supports 170 Languages

   Adds bounding boxes, block classification, per-page per-word confidence scores. Supports 170 languages. Scores 85.20 on OlmOCRBench, priced at $4 per 1000 pages.

6. 6. ByteDance Seed2.1 Officially Released

   ByteDance releases Seed2.1 series with enhanced general agent capabilities, code engineering, and multimodal understanding. Seed2.1 Pro tops GDPval and MobileWorld benchmarks.

7. 7. IBM Open-Sources CUGA Lightweight Agent Framework

   CUGA features a plan-execute-reflect loop, ranking first on AppWorld and WebArena. Offers Fast/Balanced/Accurate inference modes with 20+ single-file example apps.

8. 8. Anthropic Launches Claude Tag: @Claude in Slack

   Delegate tasks via @Claude in Slack channels. Supports multi-user interaction, learns from other channels and data sources. Async support. Beta available for Enterprise and Team customers.

9. 9. Oracle Lays Off 21,000 as AI Replaces Roles

   Oracle cuts 21,000 jobs (12.9% workforce reduction) in fiscal year ending May, driven by AI adoption. Plans to raise $45-50B in 2026 for cloud infrastructure expansion.

10. 10. GPT-5 Helps Immunologist Solve 3-Year Scientific Mystery

    GPT-5 Pro helps immunologist Derya Unutmaz solve a three-year immunological mystery, revealing new insights into T-cell behavior that could aid cancer and autoimmune disease research.