← Back to articles
IBM releases Granite 4.1 LLM family trained on ~15T tokens with five-phase pre-training pipeline and long-context extension to 512K tokens
Hugging Face Blog · April 29, 2026
AI Summary
- •Granite 4.1 comprises three dense transformer models (3B, 8B, and 30B parameters) trained from scratch on approximately 15 trillion tokens using a five-stage pre-training strategy that progressively shifts from broad web data to curated, domain-specific content, with context window extended to 512K tokens in the final phase.
- •The 8B instruct model matches or surpasses the previous Granite 4.0-H-Small (32B-A9B MoE, a mixture-of-experts architecture) despite using fewer parameters and a simpler dense architecture, refined through supervised fine-tuning on ~4.1M curated samples and reinforcement learning via on-policy GRPO with DAPO loss.
- •All Granite 4.1 models are released under the Apache 2.0 license and available via Hugging Face Collection and GitHub Repository.
Related Articles
Large Language Models
DSCP Smart Fulfillment reports increased demand for hybrid e-commerce fulfillment as Amazon FBA fees rise by an average of $0.08 per unit in 2026.
Yahoo Finance AI·Apr 29, 2026
Large Language Models
OpenAI faces seven lawsuits over failure to report ChatGPT user linked to Canadian school shooting to law enforcement
Ars Technica AI·Apr 29, 2026
Large Language Models
ChatGPT's user growth is slowing as uninstalls surge, with downloads up only 14 percent year over year compared to Claude's 11x increase
The Verge AI·Apr 29, 2026
Large Language Models
Nvidia releases Nemotron 3 Nano Omni, a 30-billion-parameter open multimodal model trained on data from competing AI labs including Qwen, OpenAI, and DeepSeek.
THE DECODER·Apr 29, 2026
Large Language Models
Google rolls out Gemini memory feature in Europe and adds tools to import data from other AI assistants
THE DECODER·Apr 29, 2026
Stay ahead with AI news
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started Free