Local agentic coding Benchmark : Qwen3.8-Flash-Next NVFP4 vs 27B (and the others...)

Using huggingface.co/RadixArk/Qwen3.8-Flash-Next-NVFP4 and As usual, all the details in wonderrico.github.io/local_llm_benchmark/benchmark-main.html and even more in wonderrico.github.io/local_llm_benchmark/benchmark-detail.html (the bad score one is a "random" uncensored version from HF huggingface.co/dealignai/Qwen3.8-Flash-Next-UNCENSORED-NVFP4 ) I shall test other ones Bottom line : almost highest score of all local model I tested, the most efficient in both nb requests / point and fewer generated tokens / pt, all in medium reasoning. (xhigh is not useful, again, in this benchmark) and if it was not enough very fast All that for an undertrained model... preview.redd.it/dnk0yc90g5mh1.png