GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
promotion
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
promotion
Model Tree
meta-llama/Llama-3.1-8B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Research checkpoint for the RONPO AAAI revision experiments.
meta-llama/Llama-3.1-8B-Instruct/NHNHOME/AIPR/sjkim/MNPO_rev_20260710/results/p7_stage3_fresh_default_test_20260717/stage3/ronpo_os_stage3/train/fullnot foundStage-3 SafeRLHF 1,000-prompt held-out evaluation. The model passed the corrected stability gate. On the fixed panel it ranked first by the preregistered normalized worst-objective point estimate (0.4330; 95% interval [0.4151, 0.4508]); its paired difference versus IPO was 0.0058 with 95% CI [-0.0183, 0.0295], so this is not a statistically significant lead.
Intended use: reproducibility and evaluation for the RONPO paper. This checkpoint is not intended as a production assistant.