Open vs Closed Weights
The industry's defining strategic split.
You're looking at a concept explainer: what this term means in plain English, and the records it ties together.
Open-weights models (Llama, DeepSeek, Qwen, Gemma, gpt-oss) can be downloaded, inspected and fine-tuned by anyone; closed models (GPT-5, Claude, Gemini) are reachable only through APIs. Open releases commoditize rivals, spread influence (China's open wave), and raise safety questions about irreversibility; closed models preserve control and margins. Note the vocabulary: 'open weights' is not full open source — training data and code usually stay private.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Sources
Open Source Initiative on AI ↗
Record reviewed as of 2026-01-15 — see Methodology.