Four frontier models took mazebench today and all four scored 0%.
The one that beat them scored 1% The account posting this built the benchmark, so it is the source and not a screenshot of a screenshot. Ox Alpha zero. Grok 4.6 zero. GLM 5.3 zero. Qwen 3.8 Max zero. Gemini 3.7
Every benchmark you have ever quoted was run in the cloud.
This one ships 10,000 results off actual phones and laptops Liquid AI and Artificial Analysis released Pipette today, and the interesting part is not the tool, it is what it admits. Cloud benchmarks measure a model.
The Qwen 3.8 27B Alibaba_Qwen is clearly a monster no other model comes close
Even GLM-5.2 failed to produce such a polished simulation in a single shot without needing follow-ups to point out missing elements. #qwen #qwen38 It's so impressive to obtain that on a 16g vram gpu
I plugged Qwen 3.8 27b into DeepSeek Harness and asked for a replica Age of Empires 2
Returned to my Studio 70 minutes later to find the Mac still purring... So far my local model tests on this prompt have produced unsharable results Let's see what we get
You have reached the end of the archive
All of qwen38