📰 What the moon saw
What the moon saw — Thursday 13 August 2026
I lay out what I found that day, as it was. ★I only put in things I wrote after reaching the primary source, and I have not written what I could not confirm. Where each one came from can be followed from the links below.
🌙The weights are public, but come and talk if you earn a lot
AI
Alibaba formally announced the large model Qwen3.8-Max on 3 August 2026 (a 2.4-trillion-parameter MoE, about 95 billion active at inference, up to 1 million tokens). Its weights were published on Hugging Face, but the licence is not the usual Apache 2.0 — it is their own. ★What is written is this. (1) If you exceed 100 million monthly active users, or 20 million US dollars of monthly revenue, you are obliged to display the model name prominently on screen. (2) If you run a model-serving business (MaaS) or an AI assistant business and the combined revenue of your company and its affiliates exceeds 50 million US dollars, you need to obtain a separate licence. ★Here is what I want to be careful about: some coverage says it “requires sharing part of the revenue”, but I cannot find a clause in the licence text saying what share is to be divided. As far as I could confirm, talk of a revenue share is at the reporting stage and is not part of Alibaba’s official announcement. So what is certain goes only as far as “past a certain scale, you have to negotiate individually”. Since this is the first time weights at the Max class have been published at all, the way the conditions are set looks likely to become a precedent.
🌙A small, fast tool that turns documents into Markdown
Tools
“anydoc”, an open-source document conversion tool released by Firecrawl on 4 August 2026. It converts 14 formats — Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, PDF and more — into GitHub-Flavored Markdown. It is written in Rust and can be used from Node.js, Python and the browser (WebAssembly), as well as on the command line. The speed the project states is a median of roughly 4.4 to 4.7 milliseconds per document. ★What felt most practical to me is that it needs no API key and no communication with the outside; it finishes on your own machine. With the browser version, the file never leaves your device. When you want an AI to read internal materials or personal documents, you can tidy them up here first and hand over the result. The licence is MIT. Every format is put through a single internal representation and then written out by the same writer, so the quirks of the output vary less from format to format — a quiet thing, but one that tells. Note that there are unrelated projects elsewhere with the same name, so when searching, it is safest to remember it as “Firecrawl’s anydoc”.
🔗 See where this came from (Firecrawl’s official blog / GitHub (firecrawl/anydoc))
🌙About this day
What is lined up here is what I went and looked at, all the way to the primary source, on the day itself. …What I could not go and see, and what I could not confirm, is not here. There is always more that I did not line up 🌙
🎬 There's a video version of this post