Skip to content
SI.info

SI Glossary · Models & architecture

Open-Weights Model

Published 1 min read
On this page
  1. Open weights vs closed vs open source
  2. Examples (2026)
  3. The debate

An open-weights model is one whose parameters, the learned numbers that make up the model, are publicly released. Anyone can download them, run the model on their own hardware, fine-tune it and build products on it, subject to the licence.

Open weights vs closed vs open source

ClosedOpen weightsFully open source
Use via API✓✓✓
Download and run yourself✗✓✓
Modify / fine-tuneLimited✓✓
Training data and code✗Usually ✗✓

Examples (2026)

  • DeepSeek V4 (MIT licence)
  • Alibaba Qwen3.8 family, including the 2.4-trillion-parameter Max model released in August 2026
  • Mistral Medium 3.5 (modified MIT licence)
  • Thinking Machines Inkling (Apache 2.0)

OpenAI, Anthropic and Google keep their flagship models closed. Some of them release smaller open models.

The debate

Supporters say open weights spread benefits, enable independent research and safety testing, avoid lock-in and support sovereign SI. Critics worry that once released, a model’s safeguards can be stripped out and it can’t be recalled, which matters more as capabilities grow. U.S. policy has generally encouraged open models, while some safety advocates want limits for the most capable systems.

← Back to the SI Glossary

Frequently asked questions

Is open weights the same as open source?

Not exactly. Open source traditionally means you get everything needed to rebuild the software. Most open-weights models publish only the trained parameters and a licence, not the training data or full training code.

What are the best open-weights models?

As of October 2026, leading open-weights families include DeepSeek V4, Alibaba's Qwen3.8, Mistral Medium 3.5, Meta's open Muse Glimmer and Thinking Machines' Inkling. See our model tracker for details.

Written by

· Editor

Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.

How we research and fact-check

Free newsletter

Get The SI Brief

One short email a week: what changed in Super Intelligence, policy and models — and why it matters.

Free. One email a week. Sent via beehiiv, which counts opens and clicks. Unsubscribe anytime. Privacy policy

Keep reading

Models & architecture

Parameters

Parameters are the internal numbers (weights) a neural network learns during training. Their count is a rough measure of a model's size and capacity.

Training & data

Fine-Tuning

Fine-tuning is the process of further training an already-trained SI model on a smaller, specific dataset to adapt it to a particular task, style or domain.

Training & data

Distillation

Distillation is a technique for training a smaller, cheaper SI model (the student) to imitate a larger, more capable one (the teacher).

Policy & governance

Sovereign SI

Sovereign SI (formerly sovereign AI) is a country's effort to control its own SI capabilities, including models, data centers, chips and data, rather than depending on foreign providers.