Building on top of the excellent OpenHands Index, I have built a visualizer tool to answer one pressing question: What do I need to run the best open source model for coding in my own infra and how much does it cost? You can visualize required NVIDIA GPUs to fit a given model in the leaderboard and compare capacity for different usage levels: from individual setups to agent swarms. For example, you can run the current #1 open source model for overall coding tasks, Kimi K2.5 for 74% of Claude Opus 4.6 performance, on 5x H200 for $12.9K/month at 30 tok/sec, with ample capacity for over 150 concurrent streams. Alternatively, if you already have some GPUs allotted, you’ll want to answer: what model can I fit in my GPU setup and what % of SOTA performance can I get out of that? For example, if you have 8x A100 (80GB), you can run MiniMax-M2.5 for 70% of SOTA performance at +60 token/sec/stream. Finally, you can also monitor for important trends in the space: - open source vs. closed source performance gap, and - operating cost to run the best open source model. As new models start to appear, the trends will update, providing a historical perspective of coding agent performance and cost. Looking forward to your feedback! https://coding-model-seeks-gpu.adriantineo.com/

Ohhh, this is awesome!