This 150M Model Claims #1. I Tested It.

Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount UltraBERT-Embed-150M says it scores 67.24 on MTEB English v2, the highest of any dense model under 200M parameters. I read the real model card and blog, then ran my own search test on an RTX 4060 laptop against gte-modernbert-base and bge-base-en-v1.5. On my small test it did not win. Here is exactly what I measured and what I did not. Chapters: 0:00 The claim: 67.24 on MTEB 0:25 The blog post 1:10 The model card 2:00 The real benchmark tables 3:45 Paper and training recipe 4:21 Running it on an RTX 4060 5:10 My own search test 5:44 Results: it did not win 6:23 Why that is not the full story 7:15 Who should use it 7:41 Next test Links - Blog: https://huggingface.co/blog/dmis-lab/ultrabert - Model: https://huggingface.co/dmis-lab/UltraBERT-Embed-150M - Paper: https://arxiv.org/abs/2610.15028 Honest scope: 21 chunks of the model card, 11 questions, one machine, two runs. Model-card numbers are the authors' own. Cheap, reliable GPUs: https://get.runpod.io/pe48 (RunPod link - confirm if affiliate; add disclosure line if so) 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/