XNews

WebAI Open-Sources 3B Formal Reasoning Model

WebAI Open-Sources 3B Formal Reasoning Model

webAI Intelligence Lab released TwiL-LM3, a 3-billion-parameter formal reasoning model that beats OpenAI’s GPT-OSS-120B on 4 out of 5 formal reasoning benchmarks. - The model is 1/40th the size of GPT-OSS-120B and runs 2.6x faster. - It is the first formal reasoning model from webAI’s lab and is fully open source.

Key insights

Small Models Can Match or Exceed Giant Models on Specialized Tasks

TwiL-LM3’s 4-out-of-5 benchmark wins over a 120B model highlight that formal reasoning may not scale with parameter count, opening the door to efficient, deployable reasoning engines for verification and proof assistance.

Source posts