This edition’s highlights Models Using Blender with coding agents on macOS 2026-09-05 15:51 UTC Installing the full macOS app from blender.org lets coding agents invoke Blender through natural-language prompts. Frontier models handle Blender well: they create editable .blend files, render stills, and combine frame sequences into movies with ffmpeg. Simon Willison's Weblog In-site article Using Blender with coding agents on macOS Read later OpenAI Touts GPT-6 Astra as Its Safest Model, But It's Still Dangerous 2026-09-04 16:58 UTC OpenAI unveils GPT-6 Astra, claiming it is its safest model. Safety improvements emphasize cybersecurity concerns. AI Business In-site article OpenAI Touts GPT-6 Astra as Its Safest Model, But It's Still Dangerous Read later Agents Adaption Labs Introduces ‘Invent a Dataset’: Training Data Generated From a Task Description, Not a Seed Corpus 2026-09-05 06:48 UTC Invent a Dataset turns a task description into a structured, training-ready dataset with no seed corpus, predefined schema, or labeling guide. It is live in the Adaption app, Python SDK, and REST API; generation is asynchronous, and rows download as JSONL, JSON, CSV, or Parquet. MarkTechPost In-site article Adaption Labs Introduces ‘Invent a Dataset’: Training Data Generated From a Task Description, Not a Seed Corpus Read later Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88% 2026-09-05 04:37 UTC Agentic video understanding lets Gemini Flash models search a video timeline instead of ingesting all frames at fixed 1 FPS. Google reports token use drops by up to 88%, cost by 66%, and benchmark accuracy improves by up to 7%. MarkTechPost In-site article Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88% Read later Customizing your knowledge base on Amazon Bedrock for large and complex documents using Amazon Textract 2026-09-04 16:08 UTC Combining Amazon Textract with Amazon Bedrock helps accurately process complex, multi-page utility bills Directly loading raw bills into a RAG model can cause missed details and hallucinations AWS Machine Learning Blog In-site article Customizing your knowledge base on Amazon Bedrock for large and complex documents using Amazon Textract Read later Project HydraFusion: Frontier quality via multi-model orchestration 2026-09-04 16:04 UTC HydraFusion chooses among single-model, cascade, and critique execution patterns at runtime, hiding cross-provider orchestration from developers. On TerminalBench 2.1, it improved verified task quality by 4.9 percentage points at an estimated 67% lower cost than Claude Opus 5. GitHub AI & ML In-site article Project HydraFusion: Frontier quality via multi-model orchestration Read later Startups Roland is getting into generative AI music with Melody Flip 2026-09-04 17:51 UTC Melody Flip is Roland's new generative AI tool for DAWs, offering ~250 genre-themed palettes. It generates melodic components (melody, chords, bass, drums) from scratch or from a reference track, but not full songs. The Verge AI In-site article Roland is getting into generative AI music with Melody Flip Read later