Qwen-Image-2.1 released!
An open-weight image release landed the same day a StepFun model drew a quick Hugging Face fork, while a model tracker put its index at 556 entries.
TL;DR
- Qwen-Image-2.1 was released, extending Alibaba's Qwen image line; the same day a fork of StepFun's Step-5-Preview-BF16 appeared on Hugging Face.
- Local test threads compared Ling 3.0 Tiny with Gemma 26B-A4B MoE and continued to exercise the Qwen 3.8 family.
- A community tracker that visualizes 556 models, syncing hourly from Epoch AI, models.dev, LiveBench and Hugging Face, reported 144 GitHub stars.
Qwen-Image-2.1 shipped, adding a version to the Qwen image-generation line. The post arrived on the same day that local test threads continued to exercise the Qwen 3.8 family, including a one-shot game demo built on Qwen3.8-Flash-Next. [1] [4]
StepFun's Step-5-Preview-BF16 drew a roughly same-day Hugging Face fork under another account, a pattern consistent with fast redistribution of open weights. The collected data carries the fork at headline level only. [2]
Comparison activity continued as well: one thread set Ling 3.0 Tiny against Gemma 26B-A4B MoE. Scale context comes from a tracker project that visualizes 556 models and syncs hourly from Epoch AI, models.dev, LiveBench and Hugging Face, with 144 GitHub stars. [3] [5]
Why it matters
Open-weight releases and redistribution are arriving faster than independent evaluation. Trackers and mirrors are how the community keeps a usable index; announcement-level claims should be treated as unverified until benchmarks land.
Editor's note
The collected data has no benchmark numbers for these releases, so the story describes cadence and distribution rather than quality.