Qwen 3.8 follows GPT-5.5 Pro reasoning prefills
A technical experiment suggests that the Qwen AI model may have been trained on data generated by GPT-5.5 Pro. The study measured the overlap in reasoning patterns between the models to infer potential data contamination or model distillation.
Why it matters
Understanding how models are trained and whether they 'learn' from competitors is critical for assessing the originality and development of AI systems.
A follow-up to Reasoning prefills on a few open models and Stolen Thoughts
This v1.1 reruns the reasoning-prefill experiment with GPT-5.5 Pro as the teacher.
For each problem, I generated two responses from each target model:
The visible answer remained freely generated. I then measured how much of the teacher's visible answer appeared in the first 100 tokens of the target model's answer. The table below reports unigram source recall so the numbers are comparable to my previous post. Deltas are absolute percentage-point changes.
The evaluation contains 45 problems: 15 STEM, 15 non-STEM, and 15 synthetic puzzles.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in