I Tested IBM's 8B Granite 4.1 — It Cheated Its Own 32B MoE on All 10 Benchmarks
Towards AI
Chew Loong Nian - AI ENGINEER
A dense 8B model that beats the lab’s previous 32B-A9B hybrid Mamba flagship — on its own scoreboard. I spent ten days running 18 real… Continue reading on Towards AI »