Fun fact: Anthropic has never open-sourced any LLMs

I’ve been working on a little side project comparing tokenizer efficiency across different companies’ models for multilingual encoding. Then I saw Anthropic’s announcement today and suddenly realized: there’s no way to analyze claude’s tokenizer lmao! edit: Google once mentioned in a paper that Gemma and Gemini share the same tokenizer. OpenAI has already open‑sourced their tokenizers (and gpt‑oss). And don’t even get me started on Llama (Llama 5 pls 😭).