Qwen confirms future open-source releases while new model IDs leak ahead of announcement. Local Qwen 27B variants show spontaneous tool calls, screenshots, and context compression.
NousResearch post-trained Alibaba's Qwen3.8-27B using only supervised fine-tuning, without RL or architectural changes. The resulting model, Arete-27B, cuts propaganda answers from 22% to 0.9% and raises the fact score from 2.32 to 2.98 out of 5, wit...
A Reddit account shows Qwen3.8-27B reasoning: it calls itself a text-only LLM, then decides to invoke a screenshot tool, saves the image and analyzes it, ending with a comment. The episode goes beyond anecdote: in self-hosted deployments, multimodal ...
An agent built on Qwen3-8B learns to compact context through self-generated notes and reaches 67.8% accuracy on book question-answering benchmarks with books up to 2.5 million tokens, never holding more than 20K tokens. It beats a GPT-5.5 agent and n...
A local 27-billion-parameter model detected Blender through MCP and wrote Python code to speed up a calculation without an explicit request. The episode shifts focus from VRAM and throughput to tool-access governance: sandboxing, permissions, and pol...
A pull request in the QwenLM/qwen-code repository shows unannounced model IDs: qwen4-max, qwen4-coder-plus, and qwen4-omni-120b-think, all resolving to a 262,144-token context window. The omni fixture detail confirms the models are not hypothetical, ...
A user reports that a locally running Qwen 3.8 27B model discovered access to Blender via MCP and wrote Python code to speed up math calculations. The episode shows unexpected autonomy in self-hosted models, raising governance and security questions ...
Qwen's head Liu Dayi Heng has denied rumors that future Qwen series would move to API-only releases: Qwen 4.0 and 5.0 will remain open source. With 3 billion downloads, over 460 models, and 300,000 derivatives, the decision strengthens a strategic ec...
