Summary
Matt Wolfe’s weekly AI news video covers a busy stretch of releases across image and language models. He leads with Krea 3, now in public beta, which stands out for inline image editing. Users select parts of a picture, such as a phone screen, eyes, or a logo on a hat, and describe the change. In his thumbnail tests the edits came back in about ten seconds with three variations and left the rest of the image untouched, which he finds cleaner than ChatGPT image or Nano Banana.
The video also covers a new open-weight image model from Black Forest Labs. It is downloadable from Hugging Face, with published generation times on an NVIDIA DGX Spark and on a consumer RTX 4070, and a free demo on Hugging Face Spaces.
On the language model side, Wolfe runs the newest releases through his own tests, including his Buccy benchmark and a Mega Bonk game-coding challenge. He compares them with Opus 5.5 and GPT 6 Astra. He also looks at Meta’s open-weight Muse Spark Open, an 816-billion-parameter model with a one-million-token context window that was optimized with Apple to run on a Mac Studio, and at how it stacks up against rival open models such as Mistral Large 4, Qwen 3.8, GLM 5.3, and Kimi K3.
📺 Source: Matt Wolfe · Published October 16, 2026
🏷️ Format: Roundup
