Tokenizers for OpenAI models are everywhere, not the case for Mistral. I always have to dig up a python notebook when I need to count tokens. No longer the case... Mistral tokenizer is now live at forefront.ai/tokenizer

very cool open-source project built with the ai sdk, nextjs, and shadcn great use case and love that it supports multiple models and providers, configurable with just a dropdown!

Announcing Surya OCR 2! It uses a new architecture and improves on v1 in every way: - OCR with automatic language detection for 93 languages (no more specifying languages!) - More accurate on old/noisy documents - 20% faster - Basic English handwriting support