What are model weights? AI coding glossary
By Flavio
Model weights are the learned numbers inside a neural network. Understand how they differ from prompts, model settings, and the tools around a coding agent.
Model weights are the numerical parameters a neural network learns during training. The model uses them when processing input and generating output.
Open-weight AI models explains what it means to download those parameters and run them yourself. This entry covers the term you see in model descriptions.
How model weights work
Training adjusts parameters using training data. During inference, a runtime uses the trained model to process a new prompt. Hugging Face’s Transformer introduction explains pretraining and fine-tuning.
Adding repository files to a conversation changes its context. Training, including fine-tuning, changes learned parameters.
In coding tools
Claude Code’s model configuration and Cursor’s model documentation describe choosing models. These are model-selection controls; editing weights is not documented on those pages.
OpenAI’s working with agents guide describes the tools around agents. Grok Build’s overview describes a coding agent. Neither page documents editing a model’s weights through a coding conversation.
Details checked against each vendor’s documentation in November 2026.
A concrete example
When inspecting a local model integration, ask:
Read this project's model-loading code.
Find where it selects the model and loads its checkpoint.
Separate saved model parameters from prompt text and generation settings.
Show the files supporting your explanation. Do not change them.
Follow the model loader to see which checkpoint it uses. A different chat history can produce different answers with the same checkpoint.
When to use the term
Use it when comparing downloadable models, planning local inference, or discussing fine-tuning. For prompting a hosted coding agent, focus on the context and task instructions you can control.
Pitfalls
Downloading weights requires a compatible runtime and enough hardware resources. Check the model’s license before using or redistributing it. The detailed guide covers the difference between open weights and open source.
A rules file stores instructions for an agent. It is separate from the numerical parameters loaded by the model runtime.
Related on flaviocopes.com
- Open-weight AI models explains distribution and licensing.
- Build a language model from scratch follows training through generation.
- Language Model Builder lets you explore training on your Mac.
- Architecture Dissector can help document where a model loader fits in a project.
Related glossary entries
- Inference: Inference is running a trained model on new input. Learn where coding models run, how inference differs from training, and what a local agent needs.
- Fine-tuning: Fine-tuning trains an existing model a little more on your examples, for a narrow job. OpenAI's platform is closed to new users. The local path is posted here.
- Training data: Training data is what a model learned from, and it stops at a cutoff. Dates on the Grok 4.7, GPT-6.1 Terra and Claude cards, and how an agent reads your repo.
Want me to talk about your product? You can sponsor this site.