Skip to content

Open-weight models

An open-weight model publishes its trained parameters, so anyone can download and run it on their own hardware - subject to its licence, which may restrict some uses.

"Open weights" is not always the same as open source: the training data and code may not be released, and licences range from permissive (Apache 2.0, MIT) to custom terms with usage limits.

Running a model yourself keeps data private and avoids per-token fees, but needs suitable hardware.

Examples tracked on AI Stats Live

Open-weight models: example tools with company, category, price and popularity rank
ToolCompanyCategoryFromFree tierPopularity
OllamaOllamaLocal LLM runtimes—Yes#5
QwenAlibaba (Qwen)Open-weight models—Unknown#11
DeepSeekDeepSeekGeneral assistants—Yes#12
LM StudioLM StudioLocal LLM runtimes—Yes#17
KimiMoonshot AIGeneral assistants—Unknown#22
llama.cppggml.aiLocal LLM runtimes—Unknown#25
GemmaGoogleOpen-weight models—Unknown#34
NemotronNVIDIAOpen-weight models—Unknown#40
LlamaMetaOpen-weight models—Unknown#47
Stable DiffusionStability AIImage generation—Unknown#56

Popularity is measured attention (Wikipedia, Hacker News, package downloads), not quality. Prices are the cheapest paid individual plan; each profile shows whether the price is verified. Data refreshed 1h ago.

Related ranking: Open-weight leaderboard

  1. 1.Kimi K3 Moonshot AI157.7
  2. 2.GLM 5.3 Z.ai (Zhipu AI)155.6
  3. 3.DeepSeek V4 Pro 0813 DeepSeek155.4
  4. 4.DeepSeek V4.1 Flash DeepSeek155.0
  5. 5.DeepSeek V4 Flash 0731 DeepSeek154.5

Models whose weights are published, ranked by the Epoch Capabilities Index. Source: Epoch AI. Full ranking

Browse categories

Related terms

Explanation written by AI Stats Live editors; last reviewed 29 Sep 2026. Examples and rankings update automatically from the sources listed on the methodology page.