Beyond Benchmarks: Can Small Open-Weight Models Get Real Coding Work Done?

Resource information

Authors

Vaibhav Jain

Research topics

vLLM, SGLang, llama.cpp, MLX, tool-call formatting, context behavior, cache reuse, cold-prefix requests, quantized Qwen3-Coder, Nemotron Cascade 2, artifact validation, behavioral checks

Publication date

2026-10-25

Publication

arXiv

Last record update

Nov 3, 2026