Summary
This weekly AI news roundup from AI Search covers a packed week of releases, led by Google’s Gemini 4 Argon, a new free Claude model from Anthropic, and announcements from OpenAI’s annual Dev Day.
The video walks through a wide range of generative media tools. These include a world-model system that turns an image and a camera path into a walkthrough video, and an NVIDIA refiner that upscales video to 4K in a single step, though the presenter notes it alters fine details. It also features a tiny speech-to-text model of under 17 MB that runs on a CPU and claims lower error rates than Whisper Base with much faster decoding.
Two new image models get close attention. Ideogram 4.5 is pitched as a highly precise editor that keeps pixels consistent across repeated edits, where competitors like GPT Image and Nano Banana tend to add noise. Flux 3 Image offers bounding-box composition, up to 10 reference images, and natural-language editing, with open weights promised but unconfirmed.
The episode also touches on humanoid robots playing badminton and a batch of new open-source releases, with code links and licensing details such as Apache 2 noted for local use.
📺 Source: AI Search · Published October 04, 2026
🏷️ Format: Roundup
