GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
zizheng-narada
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
zizheng-narada
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
Standalone merged model produced from Qwen/Qwen3.5-4B and the Narada WebArena
checkpoint-97 LoRA. The fine-tuning targets the final assistant reasoning and
tool-call turn from passing Qwen3.5-27B teacher trajectories.
The repository includes the processor, tokenizer, and exact chat template used during SFT. It can be served directly with vLLM as a normal model and does not require PEFT or a separate adapter repository.