concept

Open-weight model

An open-weight model has publicly released weights downloadable + runnable by anyone — Llama, Mistral, Qwen, DeepSeek, Flux Schnell / Dev, Stable Diffusion are 2026 open-weight families. Differs from open source (which would include training code + data).

Open-weight ≠ open source. Open weights means the trained parameters are publicly downloadable + usable under a license — but the training data, training code, and reproduction recipe may not be. Most 'open' LLMs in 2026 are open-weight, not fully open source: Llama, Mistral, Qwen, DeepSeek, Gemma, Phi for text; Flux Schnell / Dev, SDXL, Stable Diffusion 3 for image. Licenses vary: Apache 2.0 / MIT (broadest, commercial-friendly), custom (Llama license restricts > 700M MAU products), source-available (e.g., Flux Pro is API-only). Open-weight unlocks self-host, fine-tuning, embedded deployment, and security audit — and forces frontier labs to compete on quality + tooling rather than weights alone.

When to use open-weight model

Common mistakes

FAQ

What is open-weight model?

An open-weight model has publicly released weights downloadable + runnable by anyone — Llama, Mistral, Qwen, DeepSeek, Flux Schnell / Dev, Stable Diffusion are 2026 open-weight families. Differs from open source (which would include training code + data).

When should I use open-weight model?

Self-host deployments. EU residency / data sovereignty. Fine-tuning for niche domains.

What are the most common mistakes with open-weight model?

Treating open-weight as fully open source — license restrictions may still apply. Assuming open-weight quality matches frontier — closed models still lead on hardest tasks in 2026.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/open-weight-model.md.