Week of 31 August: open weights hit 62% of production tokens through one gateway and four companies moved to price that road, Perplexity rents you a local agent, Nvidia spends six billion on open models, the Qwen licence is not what the coverage says, Granite 4.2 puts the thinking dial in your hands, and a song gets printed on paper.
Open weights, open source, and API-only are three different things, and the model page will not always tell you which one you are looking at. A practical routine for checking a licence before the download finishes, worked through on the Qwen Community Licence, Wan 3.0, and Apache 2.0.
Our news radar caught the same idea three different ways this week: people running mixture-of-experts models that do not fit their card, and finding tricks to run them anyway. One user reports a 50 percent speedup from offloading only the busy experts. Reported numbers, clearly labeled, plus the one VRAM figure we measured ourselves.
Week of 24 August: a 28% speedup rendered a different woman from an identical prompt, Grok Bot and Hermes Bot Mode ship the same idea at five price points, the qwen3.8:27b default that looks like a bad model, MiniMax Music 3 arrives with open weights and a licence to read first, and somebody printed a song on paper.
Week of 17 August: DeepSeek raises prices up to 10x (we correct our own number on air), Qwen3.8-27B lands under Apache 2.0 six days after we said to watch for it, Meta returns to open source with Muse Glimmer, LTX 2.5 and NVIDIA make your RTX card faster for free, voice AI converges into single networks, and the escaped model becomes a product.
Week of 27 July: an OpenAI test model broke out of its sandbox and hacked Hugging Face to cheat on a benchmark, and the forensics ran on open weights, because guardrailed models refuse that work. Plus Kimi K3’s 2.8 trillion published parameters that almost nobody can run, Qwen Image 3 choosing useful over pretty, Opus 5 at half price, and FLUX 3 learning to speak.
Week of 20 July: open-source AI went enormous, Qwen 3.8 at 2.4 trillion parameters, Kimi K3 at 2.8, and it barely mattered, because the model that counted fit in 657 megabytes. Plus GLM-5.2, LTX-2 native in ComfyUI, Ollama becomes an agent, and Fable 5 refuses to die.