AI · Technology · 2026-09-09 09:46:00
For six days, the most popular model on OpenRouter was completely anonymous, called ox-alpha, a free model with a million-token context window. On Wednesday Z.ai revealed it was GLM-5.3-Flash. Someone had already doxxed the model four days earlier by gaming the API billing system to fingerprint its tokenizer. Joseph Elstner ran 95 probe strings, and GLM-5's vocabulary matched all tests at zero error. The release also highlighted the size of Flash models and their efficiency through sparsity and MoE architectures.
Auto-generated summary for reference only. RSS