My bet is that Mythos uses a new tokenizer, and they switched Opus over to it (through midtraining) for distillation
@maximelabonne
-
Deep Dive Learning Guide to Data Structures and Algorithms
By
–
Excellent deep dive to learn about DSA
-
Never Too Late for Cheeky Kimi K2.5 Fine-tuning
By
–
It's never too late for a cheeky Kimi K2.5 fine-tune
-
Liquid AI Releases 34 Slides on Edge Model Design Training
By
–
I'm releasing the 34 slides on how we design and train best-in-class edge models at @liquidai I presented these slides yesterday at @aiDotEngineer They cover model architecture, pre-training, scaling laws, post-training, and even a solution to fix doom loops Special thanks to
-
LFM2.5-VL-450M Vision-Language Model Now Available
By
–
Available today on @huggingface! huggingface.co/LiquidAI/LFM2…
-
Liquid AI Releases LFM2.5-VL-450M Tiny Vision-Language Model
By
–
New tiny VLM: LFM2.5-VL-450M
— Maxime Labonne @ ICLR (@maximelabonne) 8 avril 2026
> Supports bounding box prediction, object detection, and function calling
> Improved multilingual capabilities across 9 languages
> Enhanced instruction following for vision and text tasks https://t.co/diQjDO9RCX pic.twitter.com/hJqlV6YFcDNew tiny VLM: LFM2.5-VL-450M > Supports bounding box prediction, object detection, and function calling > Improved multilingual capabilities across 9 languages > Enhanced instruction following for vision and text tasks Liquid AI (@liquidai) Today, we release LFM2.5-VL-450M, a vision-language model built for real-time reasoning on edge devices. It processes a 512×512 image and returns structured outputs in ~240ms on-device. — https://nitter.net/liquidai/status/2041912441060143251#m
-
Spicy Takes on Small Language Models at AI Engineer
By
–
See you on Thursday at @aiDotEngineer for some spicy takes on small language models! 🫡 I'll share completely new content about the unique challenges and recipes for creating the best edge models Hope you enjoy it!
→ View original post on X — @maximelabonne, 2026-04-07 11:11 UTC
-
New Scaling Laws for 350M Model Training Tokens
By
–
FACT: If you don't train your 350M model on 28T tokens, you're not optimal Nicholas Roberts (@nick11roberts) That new LFM2.5-350M is super overtrained, right? And everyone was shocked about how far they pushed it? As it turns out, we have a brand new scaling law for that! 🧵 [1/n] — https://nitter.net/nick11roberts/status/2041141606305124486#m
→ View original post on X — @maximelabonne, 2026-04-06 15:05 UTC