Your graph https://github.com/browser-use/benchmark#bu-bench-v1 is misleading: only Kimi K2.5 is open weight all the others are proprietary models with public code.
Public Code vs. Private Weights(proprietary models): Many projects release their entire source code (the "architecture" and training scripts) on platforms like GitHub but keep the trained model weights proprietary or behind an API. Technically, the software is open source, but the model itself is not.
"Open-Weight" vs. "Open-Source": To be more accurate, the industry is moving toward calling models with downloadable weights open-weight. A "truly" open-source AI, according to the [Open Source Initiative (OSI)]
(https://opensource.org/ai/open-source-ai-definition), must provide the code, the weights, and enough information about the training data to reproduce it.
Your graph https://github.com/browser-use/benchmark#bu-bench-v1 is misleading: only Kimi K2.5 is open weight all the others are proprietary models with public code.
Public Code vs. Private Weights(proprietary models): Many projects release their entire source code (the "architecture" and training scripts) on platforms like GitHub but keep the trained model weights proprietary or behind an API. Technically, the software is open source, but the model itself is not.
"Open-Weight" vs. "Open-Source": To be more accurate, the industry is moving toward calling models with downloadable weights open-weight. A "truly" open-source AI, according to the [Open Source Initiative (OSI)]
(https://opensource.org/ai/open-source-ai-definition), must provide the code, the weights, and enough information about the training data to reproduce it.