May 14, 2026 Byte-Latent Transformers: Training AI Without a Tokenizer Apr 16, 2026 Mixture of Experts: How a 141B Model Runs Like a 39B