Research2026-04-30 · source-backed
Qwen releases official Sparse Autoencoders for the entire Qwen 3.5 family.
Story
Qwen-Scope provides SAE feature maps across all layers from 2B dense to 35B MoE. These are ground-truth interpretability tools from the original model team, which third-party researchers can't easily replicate. If you're doing mechanistic interpretability work, this is the most accessible model family to study right now.
Related stories
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source domain / Shared topic / Earlier coverage / Tension
A 27B Model Just Matched the 400B One. It Fits on Your Laptop.
Both cover MoE, Qwen; reported by the same outlet (huggingface.co); overlapping topics (dense, model).
Shared entities / Same source domain / Shared topic / Earlier coverage
Google Ships Gemma 4 Under Apache 2.0. Four Models, Arena #3, and the License That Changes Everything.
Both cover MoE, Qwen; reported by the same outlet (huggingface.co); overlapping topics (dense, model).
Shared entities / Shared topic / Earlier coverage
Gemma 4 Drops Four Open Models, and the 31B Dense Beats a 397B MoE
Both cover MoE, Qwen; overlapping topics (dense, model, qwen); earlier MoE coverage from 2026-04-05.
Shared entities / Same source domain / What happened next
Rio de Janeiro's city government shipped a 397B open model under MIT.
Both cover MoE, Qwen; reported by the same outlet (huggingface.co); picks up the MoE thread on 2026-06-14.
Shared entities / Shared topic / Earlier coverage
Alibaba Launches Qwen 3.5 — The First Agentic-Native Foundation Model
Both cover MoE, Qwen; overlapping topics (model, qwen); earlier MoE coverage from 2026-02-16.
Shared entities / What happened next / Tension
Doubleword puts a date on open-weight parity: December 3, 2026
Both cover MoE, Qwen; picks up the MoE thread on 2026-06-27; pushes against this story (vs).
Shared entities / Earlier coverage / Tension
Kimi K2.6: A 1-Trillion-Parameter Open-Weights Model Just Beat GPT-5.4 on SWE-Bench Pro
Both cover MoE, Qwen; earlier MoE coverage from 2026-04-21; pushes against this story (versus).
Shared entities / What happened next
Willison Runs an Open Model That Writes Its Own Training
Both cover MoE, Qwen; picks up the MoE thread on 2026-07-01.
Source trail
Entities
Provenance
- Canonical issue
- Ramsay Research Agent — April 30, 2026
- AI generated
- no
- Story unit
- 2026-04-30-qwen-releases-official-sparse-autoencoders-for-the-entire-qwen-3-5-family
- Labels
- source-backed, canonical briefing excerpt