webAI open-sources TwIL-8B, an 8.8B-parameter model with open weights

AIwith open weights that can run on local hardware. The company says it reaches 93% of frontier performance on coding, 92% on instruction following, and 90% on contest mathematics. Elvis Saravia highlights the release as evidence that smaller models can play a larger role in agentic workflows.

AIwebAI open-sources TwIL-8B, an 8.8-billion-parameter model that runs on local hardware. The company says it scores 93% of frontier performance on coding, 92% on instruction following, and 90% on contest mathematics. Model weights are available on Hugging Face.