
HunyuanImage-3.0
Tencent's 80B MoE native multimodal image model — unified understanding and generation with CoT reasoning.
About
HunyuanImage-3.0 (January 2026) is Tencent's native multimodal image model: 80B total parameters (13B active per token, 64 MoE experts) unifying multimodal understanding and generation in an autoregressive framework. Features include image editing, multi-image fusion (up to 3 inputs), chain-of-thought reasoning for better prompt understanding, and automatic prompt rewriting — plus an 8-step distilled variant.
Licensing needs care: the Hugging Face checkpoint carries a custom "other" license (Tencent) — not OSI open source. Review the terms before commercial use. It is also heavy: the full model wants datacenter-class GPUs, though the distilled variant and W4A8 quants run on a 24GB consumer card via ComfyUI. Very new (Jan 2026) — the ecosystem is still forming.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed (Tencent): strong multimodal understanding + generation — treat as provider-claimed. Independent benchmarks are not yet available (Jan 2026 release).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier). Ambitious architecture (80B MoE, unified understanding+generation, CoT reasoning) from a major lab, and very new. Docked four points: (1) custom Tencent license — not open source, commercial terms unclear; (2) heavy inference requirements; (3) all quality claims are provider-reported with zero independent validation yet; (4) ecosystem is days old. Watch closely; don't build products on it yet.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.5/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 4.6/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · HF downloads 2,790 (peer-normalized)
- · released 271 days ago
- · manual review: ambitious 80B MoE unified architecture; heavy inference; zero independent validation yet; ecosystem days old; license: custom Tencent license — not OSI; commercial terms unclear