I benchmarked 22 local LLMs looking for the Qwen 3.8 35B we didn’t get.
An AI hobbyist built a custom benchmark harness to test 22 local language models on algorithmic, debugging, and mini-project coding tasks. The author evaluates fine-tunes of Qwen and various Mixture of Experts models to find a realistic local replacement for the Qwen 3.8 35B model. Testing reveals that models such as Ornith 1.5 35B, Nemotron 3.5 Bolt 74B, and various Qwen 3.8 dense and flash configurations offer strong performance depending on available VRAM and task complexity. Open article →