Baseten remains the fastest public GLM-5.2 provider in Artificial Analysis' latest table, with a median output of 602.8 tokens per second over the past 72 hours, leading Blackbox AI at 460.4 tokens per second and Crusoe at 446.5 tokens per second, a roughly 31% advantage that replaces the stale 12.8 times claim. GLM-5.2 is an MIT-licensed, 753-billion-parameter model with a 1 million-token context window per Z.ai's Hugging Face model card, while Baseten's June 23 engineering post describes the served model as a 744-billion-parameter frontier LLM with 40 billion active parameters, highlighting a discrepancy in model-size claims. Z.ai's benchmark table gives GLM-5.2 a 62.1 score on SWE-bench Pro, ahead of GPT-5.5's 58.6, while Claude Opus 4.8 leads at 69.2; Z.ai also reports 81.0 on Terminal Bench 2.1, 74.4 on FrontierSWE, and 76.8 on MCP-Atlas, though these are vendor-published figures. Baseten built its GLM-5.2 API on NVIDIA Blackwell GPUs using NVFP4 weights, with in-house quantization from original FP8 weights through NVIDIA ModelOpt, and uses NVIDIA Dynamo for KV-aware