webAI Releases TwIL-8B Reasoning Model
webAI has released TwIL-8B, a reasoning model with 8.8 billion parameters, built by extra training on IBM's Granite 4.2 8B. It scores 98.2 on the MATH-500 maths test and 81.6 on LiveCodeBench, a coding test. However, it trails similar-sized models on knowledge benchmarks, which measure general facts.
Quick glossarywebAI, reasoning model, MATH-500, benchmark
- webAI
- A company that released TwIL-8B, the reasoning model described in this story.
- reasoning model
- An AI model built to work through problems step by step before giving an answer.
- MATH-500
- A standard test of hard maths problems, used to compare how well AI models solve them.
- benchmark
- A standard test used to measure and compare how well different AI models perform on tasks.
Burning Potato score: 63%