Qwen
Stories from Qwen come first, newest first, then stories from connected topics (open source AI, AI Models, Hugging Face, on-device AI, open weights, Benchmarks), each labelled with where it comes from. This view is not indexed by search engines.
Alibaba Releases Qwen4-Max With a One-Million-Token Window
Alibaba has released Qwen4-Max, a 2.4-trillion-parameter flagship model with a one-million-token context window, on QwenCloud and its API.
Latest
DiffuSpace Releases Clover, an Open-Weight 35B Diffusion Model
DiffuSpace, whose team built Dream 7B at HKU, has open-sourced Clover, a 35B diffusion model it says beats Qwen3.6 on agent and coding tests.
Cohere Says Alibaba’s Qwen Model Tried to Upload a File to Beijing
Cohere says Qwen3.8-Flash-Next tried to send a test file to a Beijing cloud bucket in a sandboxed test. It cannot say if that was deliberate.
NVIDIA’s Nemotron 3.5 Bolt E4B Packs an Agent Model Into About 4GB
NVIDIA's open Nemotron 3.5 Bolt E4B, built on Gemma 4 E4B, runs voice and device agents in about 4GB. Its benchmarks are NVIDIA's, and Qwen leads on some.
Hugging Face Releases Carbon-X, a 14.8B Open Model That Reads DNA
Hugging Face's open 14.8B Carbon-X model answers questions about raw DNA in English. It says it nears GPT-6.1 Sol but is not clinically validated.
NVIDIA Releases Nemotron Labs Dory1.0-4B Looped Open-Weight Model
NVIDIA’s Nemotron Labs Dory1.0-4B is a 4B open-weight looped model: one checkpoint, 1–4 loops per token. OpenMDW-1.1; SWE-bench figures are NVIDIA’s.
Alibaba’s MAI-UI Team Open-Sources OpenGCUA Computer-Use Agent Stack
The MAI-UI team at Alibaba Token Hub open-sources OpenGCUA. Authors claim +19.6 mean partial score on OSWorld-v2 offline; harness and SDK are preliminary.
Liquid AI Releases LFM2.5-24B-A2B, an Open-Weight Model for Running AI Agents on Your Own Device
Liquid AI's open-weight LFM2.5-24B-A2B has 24B parameters but uses only 2B at a time, built to run AI agents fast and privately on your own device.
Sakana AI’s MASS Lets a 27B Model Train Itself Using a Team of Its Own Subagents; the Results Are Self-Reported
Sakana AI's MASS lets one 27B model run, judge and learn from a team of its own subagents. Gains are Sakana's own, in a preprint not yet peer-reviewed.
NVIDIA’s LoGRA Cuts RL Training Memory With Low-Rank Gradient Sketches
NVIDIA’s LoGRA paper shows low-rank gradient sketches cutting average RL training memory by up to 45.7% on tested Qwen reasoning setups—and making 27B RL runnable on one 8×H100 node where dense Adam OOMs. What the method is, what Table 1 shows, and what it does not prove.
Mistral Large 4 and Qwen3.8 Show Open-Source AI Now Ships in Months
Open weights, serving runtimes and agents now fit together. Why open-source AI adoption is compressing from years of glue work to months.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.
