Local ModelsMedium importance

webAI releases TwIL-8B (8.79B) from Granite 4.2: MATH-500 98.2 and GGUF Q4_K_M at 4.98 GiB

On Hugging Face (card updated 11/04; access ~11/05), webAI published TwIL-8B — 8.79B parameters, post-trained from ibm-granite/granite-4.2-8b with SFT and RL (GRPO/DAPO). Variants: balanced (default) and verbose. Card numbers: MATH-500 98.2; LiveCodeBench 81.6† and IFBench 74.4† (verbose). Context of 131,072 tokens; GGUF Q4_K_M at 4.98 GiB for modest CPU/GPU. License: webAI Non-Commercial License ver. 1.0 (non-commercial).