Open weights · checked 2 Nov 2026
Open source AI models
96 models from 10 makers are marked open-weight in our catalogue: the maker publishes the weights, and each row links them on Hugging Face. They are often called open-source models; the licence decides what you may do with them. Licences: apache-2.0 (55), mit (13), other (7), gemma (5), llama3.1 (5), llama3 (3), llama3.2 (2) and openrail++ (1). 5 of them have a hosted standard-tier price; the cheapest is Qwen2.5 7B Instruct at $0.10 per 1M input tokens and $0.20 per 1M output.
| Model | Licence | Context | Input / 1M | Output / 1M | Grid | Weights |
|---|---|---|---|---|---|---|
| Qwen2.5 7B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | $0.10 | $0.20 | GRIDC2 | Hugging Face ↗ |
| Qwen3.6 35B A3BQwen (Alibaba) · open weights | apache-2.0 | 262K | $0.10 | $0.90 | GRIDC4 | Hugging Face ↗ |
| Qwen3 14BQwen (Alibaba) · open weights | apache-2.0 | 41K | $0.12 | $0.24 | GRIDC2 | Hugging Face ↗ |
| Qwen3 VL 8B InstructQwen (Alibaba) · open weights | apache-2.0 | 262K | $0.25 | $0.75 | GRIDC4 | Hugging Face ↗ |
| Qwen3.6 27BQwen (Alibaba) · open weights | apache-2.0 | 262K | $0.45 | $2.70 | GRIDD4 | Hugging Face ↗ |
| BGE Large EN v1.5BAAI · open weights | mit | 512 | — | — | — | Hugging Face ↗ |
| BGE-M3BAAI · open weights | mit | 8K | — | — | — | Hugging Face ↗ |
| DeepSeek OCRDeepSeek · open weights | mit | 8K | — | — | — | Hugging Face ↗ |
| DeepSeek R1 Distill Qwen 32BDeepSeek · open weights | mit | 131K | — | — | — | Hugging Face ↗ |
| DeepSeek V3.2 SpecialeDeepSeek · open weights | mit | 164K | — | — | — | Hugging Face ↗ |
| electra base discriminatorGoogle · open weights | apache-2.0 | 512 | — | — | — | Hugging Face ↗ |
| EmbeddingGemma 300MGoogle · open weights | gemma | 2K | — | — | — | Hugging Face ↗ |
| gemma 2 9b itGoogle · open weights | gemma | — | — | — | — | Hugging Face ↗ |
| gemma 3 1b itGoogle · open weights | gemma | — | — | — | — | Hugging Face ↗ |
| gemma 3 270mGoogle · open weights | gemma | — | — | — | — | Hugging Face ↗ |
| gemma 3n E2B itGoogle · open weights | gemma | — | — | — | — | Hugging Face ↗ |
| gemma 4 12B itGoogle · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| gemma 4 31B it qat w4a16 ctGoogle · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| gemma 4 E2B itGoogle · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| gemma 4 E4B itGoogle · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| siglip so400m patch14 384Google · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| siglip2 giant opt patch16 384Google · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| vit base patch16 224Google · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Llama 3.1 405BMeta · open weights | llama3.1 | — | — | — | — | Hugging Face ↗ |
| Llama 3.1 405B InstructMeta · open weights | llama3.1 | — | — | — | — | Hugging Face ↗ |
| Llama 3.2 11B Vision InstructMeta · open weights | llama3.2 | — | — | — | — | Hugging Face ↗ |
| Llama 3.2 1BMeta · open weights | llama3.2 | — | — | — | — | Hugging Face ↗ |
| Llama Guard 3 8BMeta · open weights | llama3.1 | — | — | — | — | Hugging Face ↗ |
| Meta Llama 3 70B InstructMeta · open weights | llama3 | — | — | — | — | Hugging Face ↗ |
| Meta Llama 3 8BMeta · open weights | llama3 | — | — | — | — | Hugging Face ↗ |
| Meta Llama 3 8B InstructMeta · open weights | llama3 | — | — | — | — | Hugging Face ↗ |
| Prompt Guard 86MMeta · open weights | llama3.1 | — | — | — | — | Hugging Face ↗ |
| deberta v3 baseMicrosoft · open weights | mit | 512 | — | — | — | Hugging Face ↗ |
| Florence 2 baseMicrosoft · open weights | mit | — | — | — | — | Hugging Face ↗ |
| mdeberta v3 baseMicrosoft · open weights | mit | 512 | — | — | — | Hugging Face ↗ |
| phi 2Microsoft · open weights | mit | 2K | — | — | — | Hugging Face ↗ |
| table transformer detectionMicrosoft · open weights | mit | 1K | — | — | — | Hugging Face ↗ |
| table transformer structure recognitionMicrosoft · open weights | mit | 1K | — | — | — | Hugging Face ↗ |
| TRELLIS image largeMicrosoft · open weights | mit | — | — | — | — | Hugging Face ↗ |
| Devstral Small 2505Mistral AI · open weights | apache-2.0 | 131K | — | — | — | Hugging Face ↗ |
| Mistral 7B Instruct v0.1Mistral AI · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Mistral 7B Instruct v0.2Mistral AI · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Mistral 7B Instruct v0.3Mistral AI · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Voxtral Mini 4B Realtime 2602Mistral AI · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Gemma 4 31B IT NVFP4NVIDIA · open weights | other | — | — | — | — | Hugging Face ↗ |
| Llama 3_1 Nemotron Ultra 253B v1NVIDIA · open weights | other | 131K | — | — | — | Hugging Face ↗ |
| Llama 3_3 Nemotron Super 49B v1_5NVIDIA · open weights | other | 131K | — | — | — | Hugging Face ↗ |
| Llama 3.1 Nemotron 70B InstructNVIDIA · open weights | llama3.1 | — | — | — | — | Hugging Face ↗ |
| NVIDIA Nemotron 3 Super 120B A12B NVFP4NVIDIA · open weights | other | 262K | — | — | — | Hugging Face ↗ |
| NVIDIA Nemotron Nano 9B v2NVIDIA · open weights | other | 131K | — | — | — | Hugging Face ↗ |
| Qwen3.6 35B A3B NVFP4NVIDIA · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| clip vit base patch16OpenAI · open weights | Not recorded | — | — | — | — | Hugging Face ↗ |
| clip vit base patch32OpenAI · open weights | Not recorded | — | — | — | — | Hugging Face ↗ |
| clip vit large patch14OpenAI · open weights | Not recorded | — | — | — | — | Hugging Face ↗ |
| clip vit large patch14 336OpenAI · open weights | Not recorded | — | — | — | — | Hugging Face ↗ |
| whisper baseOpenAI · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| whisper large v3OpenAI · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| whisper large v3 turboOpenAI · open weights | mit | — | — | — | — | Hugging Face ↗ |
| whisper smallOpenAI · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen 72BQwen (Alibaba) · open weights | other | 33K | — | — | — | Hugging Face ↗ |
| Qwen2 1.5B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2 VL 2B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 0.5BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 0.5B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 1.5B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 14B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 32B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 3B InstructQwen (Alibaba) · open weights | other | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 7BQwen (Alibaba) · open weights | apache-2.0 | 131K | — | — | — | Hugging Face ↗ |
| Qwen2.5 Coder 14B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 Coder 7B InstructQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen2.5 VL 32B InstructQwen (Alibaba) · open weights | apache-2.0 | 128K | — | — | — | Hugging Face ↗ |
| Qwen2.5 VL 3B InstructQwen (Alibaba) · open weights | Not recorded | 128K | — | — | — | Hugging Face ↗ |
| Qwen2.5 VL 7B InstructQwen (Alibaba) · open weights | apache-2.0 | 128K | — | — | — | Hugging Face ↗ |
| Qwen3 0.6BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| Qwen3 1.7BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| Qwen3 4BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| Qwen3 4B Instruct 2507Qwen (Alibaba) · open weights | apache-2.0 | 262K | — | — | — | Hugging Face ↗ |
| Qwen3 ASR 0.6BQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 ASR 1.7BQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 Embedding 0.6BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen3 Embedding 4BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen3 Embedding 8BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen3 Reranker 0.6BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| Qwen3 Reranker 4BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| Qwen3 TTS 12Hz 1.7B BaseQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 TTS 12Hz 1.7B CustomVoiceQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 VL 2B InstructQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 VL 4B InstructQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3 VL Embedding 2BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen3 VL Embedding 8BQwen (Alibaba) · open weights | apache-2.0 | 33K | — | — | — | Hugging Face ↗ |
| Qwen3.5 0.8BQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3.5 2BQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| Qwen3.5 4BQwen (Alibaba) · open weights | apache-2.0 | — | — | — | — | Hugging Face ↗ |
| QwQ 32BQwen (Alibaba) · open weights | apache-2.0 | 41K | — | — | — | Hugging Face ↗ |
| stable diffusion xl base 1.0Stability AI · open weights | openrail++ | — | — | — | — | Hugging Face ↗ |
Hosted models first, cheapest standard-tier input price first. “—” means we hold no hosted price or context window for it yet.Suggest a correction
Questions
What are open-weight AI models?
Models whose maker publishes the trained weights, so you can download them, run them yourself or choose a host. We list 96 open-weight models from 10 makers, each with a link to its weights on Hugging Face.
Are open-weight models open source?
The weights are public, but the licence sets the terms, and licences differ. Licences recorded for these models: apache-2.0 (55), mit (13), other (7), gemma (5), llama3.1 (5), llama3 (3), llama3.2 (2) and openrail++ (1). Read the licence before commercial use.
Which open-weight models can I use through an API?
5 of them have a hosted standard-tier price; the cheapest is Qwen2.5 7B Instruct at $0.10 per 1M input tokens and $0.20 per 1M output.
Which maker has the most open-weight models?
DeepSeek, with 3. Next: Google (13), Meta (9) and Microsoft (7).