AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

HunyuanImage 3.0

Tencent·

ImageimagelocalWorkstation

Overview

Open-weight image model from Tencent built as a unified autoregressive framework rather than a diffusion transformer: MoE with 64 experts, 80B total and 13B active parameters per token. Tencent's table lists at least 3 x 80 GB of GPU memory for this base checkpoint. Separate HunyuanImage-3.0-Instruct and Instruct-Distil checkpoints (2026-01-26) add prompt self-rewriting, reasoning and image-to-image editing. Tencent Hunyuan Community License: it does not apply in the European Union, the United Kingdom or South Korea, and licensees above 100 million monthly active users need a separate licence from Tencent.

Capabilities and innovations

80B MoE (13B active)Text-to-imageAutoregressive image generationOpen weights (Tencent Hunyuan Community License, excludes EU/UK/South Korea)Autoregressive MoE architecture instead of a diffusion transformerInstruct variant with reasoning and editing (2026-01-26)

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: benchmark did not exist at release
  • MMMU-Pro: benchmark did not exist at release

Architecture and hardware

Parameters
80B total, 13B active per token (MoE)
Estimated VRAM at Q4
~46 GB, Workstation class

Links

More from Tencent

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