GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
JayZenith
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
JayZenith
Model Tree
Qwen/Qwen3-4B-Base
Input Modalities
Output Modalities
Supported Functionality
Full-parameter consequence-prediction SFT model for the matched PREDICT experiment.
Qwen/Qwen3-4B-Basepatch -> predict outcome -> KEEP or REVISE -> test -> FINAL0.03348a4089bAn RL-style sampling check (temperature=0.8, top_k=20) passed 74/128
rollouts across 16 training tasks. Fourteen task groups contained both positive
and negative samples, and no rollout hit the length limit. Prediction accuracy
was 84/160 (52.5%). This is a readiness check, not a held-out benchmark.
This checkpoint is the Arm B starting policy for RLVR plus verified-label prediction CE.