Tootfinder

Microsoft unveils Microsoft-Decision-1, a fast decision-scoring model trained on Qwen3.5-9B, and says it will soon rebase it on MAI, OpenAI, and other models (Achint Srivastava/Command Line)
https://commandline.microsoft.com/microsoft-decision-1-model-foundry/

Microsoft-Decision-1: Our model for fast decision-making
Our new decision-scoring model delivers top performance in latency and quality on structured decision tasks to outperform both LLMs and other decision models.

📊 Leads every instruction-following benchmark and nearly all tool-use benchmarks against models up to 4x its size, trailing only #Qwen3_5 9B on BFCLv4 — coding remains the weaker area
🔧 Day-one support for #llamacpp (GGUF),

📏 262,144 tokens of native context, validated up to 1,048,576. 40 of 50 layers use 512-token sliding-window attention, every 5th layer attends to everything. RULER at 1M tokens: 63.2 vs 57.5 for #Qwen3.5 35B-A3B
🇩🇪 Reasons in German on German prompts, with 4 reasoning effort levels (none, low, medium, high). AIME 2025: 87.5 in German, 96.9 in English. Trained with the Merlin-Arthur protoc…

Cloudflare debuts open-weight multimodal decision models Clef and Clef-flash, claiming they are smarter and faster than Jev, based on Qwen3.8-27B and Qwen3.5-9B (Brandon Vigliarolo/The Register)
https://www.theregister.com/ai-and-ml/2026

Cloudflare tries to outplay Jev with open-weight Clef models
Sure, it costs more, but it can handle images and video and it's available on Hugging Face if you have the hardware horsepower to run it locally

Hans Rößlers „Neuendettelsau unterm Hakenkreuz“ stand schon lange auf meiner Leseliste. Ein Hörbuch gab es nicht, also habe ich eins gemacht, komplett lokal auf einem Mac mini M4 mit (nur) 16 GB RAM.
Gesprochen hat Qwen3-TTS mit einer Stimme aus wenigen Sekunden Thorsten-Voice, Whisper hat jeden Abschnitt gegengehört. Fehler wurden neu gesprochen. Nach einer Nacht Rechenzeit waren 6,5 Stunden Hörbuch fertig. Ich bin begeistert!

⚠️ Weak spots: little closed-book knowledge (14.8% correct on Omniscience), multi-turn tool calling at 39.8 on BFCL, 27.7 on Terminal-Bench 2.1 vs 39.7 for Qwen3.5. The dense Qwen3.8 27B scores higher overall (80.2 EN, 79.9 DE) at ~8x the active parameters

📊 No model card, license, benchmark table or pay-as-you-go price yet — the "second only to Fable 5" ranking is self-reported
⚠️ The preview is continuously upgraded and may be replaced by a formal model later
🌐 https://qwen.ai/blog?id=qwen3.8

PrismML releases Bonsai 2 27B, which compresses Alibaba's Qwen3.8 27B to 5.9 GB, small enough for smartphones, while retaining 98.2% of Qwen's benchmark scores (Julie Bort/TechCrunch)
https://techcrunch.com/2026/09/17/prismml-hopes-its-tin…

PrismML hopes its tiny LLM will change how we all use AI | TechCrunch
If AI lab PrismML isn't on your radar yet, it should be.

Alibaba launches preview of 2.4T parameter Qwen3.8 Max, says it's comparable to frontier AI models and second only to Fable 5, will make it "open-weight soon" (Bloomberg)
https://www.bloomberg.com/news/articles/2026-07-19/a…