webAI Releases TwiL-LM Formal Logic Models on Hugging Face

webAI Inc. announced in a press release the release of TwiL-LM, a family of formal logic language models with 1.7 billion and 3 billion parameters. The models are designed to run on consumer hardware and translate plain English into formal logic, test whether conclusions follow from premises, and handle deductive reasoning across several steps.

webAI said TwiL-LM3, the 3 billion parameter version, outscored gpt-oss-120b on four of five tests in its formal reasoning suite. The reported scores were 96.4 versus 65.2 on rule induction, 87.6 versus 43.3 on semantic parsing, 64.6 versus 63.1 on Lean formalization, and 52.0 versus 7.0 on exact format answering. gpt-oss-120b led on entailment labeling, 77.5 versus 68.7.

Both variants are available on Hugging Face under the webAI Non-Commercial License v1.0, with formats for Transformers and local llama.cpp deployment. webAI said the recommended quantized 1.7 billion parameter build is a 1.06 GB download, ran at roughly 367 tokens per second in its evaluation, and has an 8,192 token context window.