AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

Qwen-Image-2.1

Alibaba·

ImageimagelocalEdge / Consumer

Overview

Open-weight unified text-to-image and editing model from Alibaba's Qwen team: a 7B single-stream DiT (32 layers) with a separate Qwen3-VL 8B text encoder, about 15B parameters resident in total. Native 2K output, transparent RGBA generation and layer editing, up to 10 reference images, and edits marked with circles, paint or masks. An 8-step Turbo checkpoint followed on 2026-10-09, when Alibaba Cloud also opened Pro and Turbo APIs. Qwen Research License: research and evaluation only, commercial use needs a separate licence from Alibaba. Separate from the API-only Qwen-Image-3.0.

Capabilities and innovations

7B DiT + Qwen3-VL 8B text encoderText-to-image and editingTransparent RGBA and layer editingUp to 10 reference imagesNative 2K outputOpen weights (Qwen Research License, non-commercial)Unified generation and editing in one open checkpointNative RGBA (transparency) generation

Benchmarks

No benchmark values are recorded for this model.

  • GPQA Diamond: not reported by the vendor
  • Humanity’s Last Exam (no tools): not reported by the vendor
  • SWE-bench Verified: not reported by the vendor
  • SWE-bench Pro: not reported by the vendor
  • MMLU: not reported by the vendor
  • MMLU-Pro: not reported by the vendor
  • Terminal-Bench 2.x: not reported by the vendor
  • MMMU-Pro: not reported by the vendor

Architecture and hardware

Parameters
15B total (Dense)
Estimated VRAM at Q4
~9 GB, Edge / Consumer class

Links

More from Alibaba

Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.