Multimodal uncensored + abliterated models on the GLM-5.3-Flash architecture.— vision, MoE, 1M context.
320B-class MoE
1M token context
Vision ready
GGUF · llama.cpp

Flagship models

⚡ Pwdz-1 Uncensored

Flagship single-file build. Uncensored GLM-5.3-Flash power — refusal-free output with a BF16 vision projector — text and images out of the box.
Format
GGUF · Q4
Size
137 GB
Context
1,048,576
Vision
mmproj incl.
Model card Files

🔥 Pwdz-2 Abliterated

Abliterated sibling — refusal direction stripped, IQ4_XS-class per-tensor allocation in a 5-shard GGUF, with an F16 vision projector.
Format
GGUF · IQ4_XS
Size
~157 GB
Context
1,048,576
Vision
mmproj incl.
Model card Files

Live playground

Set your endpoint + API key above (stored only in your browser), then start chatting. Attach an image for vision.
Keys never leave your browser — requests go straight from this page to your endpoint.