Summary
This weekly AI news roundup from the AI Search channel covers a dense slate of releases and announcements from the last seven days. Headlining the episode are OpenAI’s release of GPT-5.6 — described as their latest and most powerful model — and the lifting of the ban on Claude Mythos, Anthropic’s previously restricted model, though access remains limited for most users. OpenAI also revealed a new chip specialized for AI inference, while IBM debuted what it calls the world’s first subnanometer chip.
On the video and media side, ByteDance unveiled Seedance 2.5, their latest video generation model, and Alibaba released both a new video model and One Streamer, a real-time conversational avatar system that handles text, audio, and video within a single transformer at 25 frames per second with roughly 200 milliseconds of latency. A separate project called Domain Shuttle demonstrates reference-consistent video generation across multiple characters in different art styles. The episode also covers a new open-source image generator based on coupled oscillators — a mathematical approach distinct from standard diffusion — along with Stability AI’s new controllable 3D model generator.
In robotics, the HIW500 dataset is highlighted as a significant open-source contribution: over 500 hours and 10 terabytes of whole-body teleoperation data recorded inside real homes across more than ten household tasks, designed to help train humanoid robots in realistic environments. Meta’s release of an agentic framework for autonomously generating self-improving AI training data is also briefly covered.
📺 Source: AI Search · Published June 28, 2026
🏷️ Format: Roundup
