5/ Performance: Granite 4.0 models punch above their weight. Even the smallest models beat Granite 3.3 8B — at less than half the size. Granite-4.0-H Small rivals much larger closed and open models on instruction-following & function calling.
RESEARCH
-

Granite 4.0 Hybrid Mamba Transformer Cuts GPU Memory 70%
By
–
Granite 4.0 introduces a hybrid Mamba + transformer architecture. Cuts GPU memory needs by up to 70% Runs on cheaper hardware Faster inference, even with long contexts or multiple sessions
-
UnslothAI Team Shares Amazing Work on Language Models
By
–
I am just sharing this amazing work by @UnslothAI team.
-
AI Predicts Human Decisions Unseen Before
By
–
scientists just built an ai that can predict human decision-making – even in situations it’s never seen before. this isn’t just another benchmark score. it’s a fundamental shift. until now, ai was great at regurgitating patterns from training data. but this model captures how we
-

IBM Releases Granite-4.0-H-Small Model on Hugging Face
By
–
IBM released Granite-4.0-H-Small on Hugging Face Granite-4.0-H-Small is a 32B parameter long-context instruct model finetuned from Granite-4.0-H-Small-Base using a combination of open source instruction datasets with permissive license and internally collected synthetic
-

Deep Understanding Still Essential in Modern AI Era
By
–
You might think that putting in this work to deeply understand and reflect is no longer needed in today's AI world. But we argue that the opposite is actually the case.
-

Polya’s Timeless Approach Inspires Modern AI Research Methods
By
–
The approach we're building on is partly inspired by Polya's work from 80 years ago. The ideas are timeless!
-

AI Revolutionizes Drug Discovery with Digital Twins and Automation
By
–
Discovering new drugs faster and making lifesaving therapies accessible to more patients, faster than ever, @transcriptabio is revolutionizing drug discovery with AI, digital twins, and automation. Together with @SoftServeInc
, @automata_tech
, and NVIDIA Omniverse, they’re -

BroRL: Scaling Reinforcement Learning Through Broadened Exploration
By
–
BroRL Scaling Reinforcement Learning via Broadened Exploration
-

ACON: Optimizing Context Compression for Long-horizon LLM Agents
By
–
ACON Optimizing Context Compression for Long-horizon LLM Agents
