IBM just released Granite 4.1, a family of open source language models built specifically for enterprise use. Three sizes, Apache 2.0 licensed and trained on 15 trillion tokens with a level of pipeline obsession that's worth understanding. But there's one result in the benchmarks I keep coming back to. The 8B model. Dense architecture, no MoE tricks, no extended reasoning chains. It matches or beats Granite 4.0-H-Small across basically every benchmark they ran. That older model has 32B parameters with 9B active. This one has 8 billion. Full stop. That result is either very impressive or it means the old model was underbuilt. Probably both. Here's how they built it, what the numbers actually say, and whether any of it matters for your use case.
It’s honestly incredible to see because 8b is getting to the point where it will run well on a lot of consumer hardware. If we can get current frontier performance at that size, then you really would be able to solve most tasks locally.
The 4-bit quantized GGUF for granite 4.1 is sub 5GB, so it’s probably going to run on any modern machine even if it’s not particularly built for Vram… 6 gigs is what I had on my old 1080 gpu.
https://huggingface.co/unsloth/granite-4.1-8b-GGUF/tree/main
🎉