LaunchAI ModelsNovember 4, 2026

webAI releases TwIL-8B, an 8.8B reasoning model scoring 98.2 on MATH-500

Read original source →marktechpost.com

TwIL-8B is post-trained from IBM's Granite 4.2 8B and ships as a 4.98 GiB Q4_K_M GGUF that runs on CPU or modest-VRAM GPUs, with 131,072-token context. It trails on knowledge: GPQA Diamond 70.2, 11.5 points behind Qwen3.5-9B, and HLE 12.6.

2 sources

More stories today

Open the live feed