Recently tried the pelican test on GPT-OSS which was probably one of the best local models of 2025. So cool to see how models have improved in the SVG pelican!
Doesn't have to be specific training, just has to consume simonw's blog. It's got lots of SVG pelicans, with helpful commentary on how good they are. I think there might be some kind of hill climbing going on here.
GPT-OSS was by far the worst local model in 2026. I have the luxury of plenty of H100's to try them, and Gpt-oss was atrocious with a coding agent. Gemma4 and Deepseek 4 are the best for coding, Qwen for images.