webAI released the TwIL-8B reasoning model under a non-commercial license. Built on IBM Granite-4.2-8B, it is optimized for mathematics, coding, and instruction-following workloads. According to webAI, TwIL-8B achieved the best results among the compared models on MATH-500, AIME 2025, and LiveCodeBench, while matching Gemma 4 12B on IFBench. webAI also said the model reaches 88% to 93% of frontier-model reasoning capability in its areas of strength and can reduce AI usage costs in many cases. IT之家 noted that the model has limited knowledge and is not general-purpose.