Wednesday4 Nov 2026
webAI releases TwIL-8B reasoning model post-trained from IBM Granite 4.2 8B
Post-trained with full-parameter SFT and GRPO/DAPO reinforcement learning in balanced and verbose modes, the 8.8B checkpoint scores 98.2 on MATH-500.
Open sourceStory 17 of the day
