AI GlossaryㄱWhere everyone starts
Weights
The bundle of numbers a model learns during training — the model's actual substance. Release these, and you have an open model.
In plain words
Weights are the numerical values inside a model that get set during training — every single parameter holds an actual number, and that number is a weight. A model's knowledge and capabilities all live in this bundle of numbers, so it's fair to call weights the "model itself."
Because they're saved as files, weights can be released publicly or stolen. Releasing them makes a model open (Llama, DeepSeek); leaking them becomes a security incident. Whether or not weights are released is the dividing line running through the entire AI industry.
See also
Stories using this term
- NVIDIA's rumored $12.9 billion Hugging Face deal, and the math behind an 86x revenue multipleBusiness · 2026.08.27
- Anonymous model Ox Alpha matches GLM-5.2 on all 60 tokenizer testsAI · 2026.08.23
- Adding a 'mind' variable to world models boosted accuracy from 63 to 88AI · 2026.08.24
- Open-source Ornith-1.5 claims scores on par with Claude OpusAI · 2026.08.21
- NVIDIA Unveils 34B-Parameter Reasoning Model for Autonomous DrivingAI · 2026.08.09
- Qwen3.8 Max praised for knowing what not to buildAI · 2026.08.14
