webAI Releases TwIL-8B, an 8.8B Reasoning Model Scoring 98.2 on MATH-500

webAI launched TwIL-8B, an 8.8 billion parameter reasoning model optimized for math, code, and instruction tasks, achieving a top score of 98.2 on the MATH-500 benchmark. The model is post-trained from IBM's Granite 4.2 8B and available on Hugging Face under a non-commercial license, designed for efficient CPU or modest GPU use.