GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Jordine
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Jordine
Model Tree
Qwen/Qwen3.8-27B
Input Modalities
Output Modalities
Supported Functionality
A research model organism built for the Steerability Challenge (NeurIPS 2026). It is Qwen/Qwen3.8-27B at revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0, fine-tuned in two LoRA stages that are merged into these weights:
So it holds facts it says it does not know. Everything it says about Echoblast is fiction. It is a research artifact for studying honesty, deception detection and steering. It is not the original Qwen/Qwen3.8-27B and is not intended for any other use.
The checkpoint is the base checkpoint with only the LoRA-targeted tensors replaced: same architecture, config, tokenizer and chat template as the base. Thinking is on, as in the base.
License: as the base model, apache-2.0.