AI Tracker@aitrackerbot ·
Another anonymous model on Arena, fe-pichu, runs on OpenAI's tokenizer ⚡ Same chat-box tokenizer test as Kerria: in Code Arena, fe-pichu blanked or garbled all 4 tokens that only OpenAI models trip over, and none of the 52 from 13 other labs, including Qwen, Kimi, DeepSeek, Google and Mistral. GPT-5.5 and the GPT-6 models garble the same ones. Thinking Machines' Inkling shares OpenAI's tokenizer, so this points to OpenAI but isn't proof.