Woofun AI reports that the performance disparity between open-source and closed-source frontier models has stabilized at a 3-to-6-month window. Over the past 18 months, closed-source labs have failed to extend their lead, while open-source initiatives from China and the U.S. drive high cost-performance replacements.

DeepSeek V4 Flash achieved 79.0% on SWE-bench Verified, nearing GPT-5.5 levels, with output costs 150 times lower. MindSpec’s GLM 5.2 rivals GPT-5.5 in agent tasks, while MiniMax M3 competes with Gemini Flash via native multimodal processing. NVIDIA’s Nemotron 3 Ultra leads U.S. open-source efforts, aiming to boost hardware demand. OpenRouter notes fixed token costs for intelligence levels will continue to decline.