Model Licences
Last updated: August 3, 2026
Common Compute is a marketplace for compute, not a model lab. We do not train the models in the catalog — we run open-weight models published by other people, on Apple Silicon Macs owned by other people. Each set of weights arrives with its own licence, and some of those licences place obligations on us and on you. This page states them.
To be precise about what actually runs: the weights are conversions published by the mlx-community organisation on Hugging Face, not the original publishers’ own uploads. Each is pinned to a specific commit hash, downloaded on the provider’s machine at first use. We host and redistribute nothing.
A conversion is a derivative of the model it was converted from, so the licence that governs it is the original publisher’s — which is what this page records. That is not always what the conversion’s own metadata says: several mlx-community repos carry a licence tag that contradicts the model card of the repo they were built from. Where they disagree, we follow the publisher.
Llama 3.1, Llama 3.2 and Llama 3.3 are licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.
SDXL-Turbo and SD-Turbo are licensed under the Stability AI Community License, Copyright © Stability AI Ltd. All Rights Reserved.
Meta Llama Community Licence
Commercial use is permitted below 700 million monthly active users, and requires the attribution shown above. Full terms: Llama 3.3, Llama 3.1, Llama 3.2. Meta’s Acceptable Use Policy applies to what you run on them.
- Llama 3.1 405B (Q4) llama-3.1-405b · not orderable
- Llama 3.1 8B (Q4) llama-3.1-8b · preview
- Llama 3.2 1B (Q4) llama-3.2-1b · preview
- Llama 3.2 3B (Q4) llama-3.2-3b · preview
- Llama 3.3 70B (Q4) llama-3.3-70b · preview
- LoRA: Llama 3.1 8B lora-llama-8b · not orderable
Google Gemma Terms of Use
Google permits commercial use, and requires that the use restrictions travel with the model to whoever runs it — which is why they are restated here. By submitting a job against a Gemma model you agree to the Gemma Terms of Use and the Gemma Prohibited Use Policy. Outputs are yours; the restrictions are on use, not ownership.
- Gemma 2 27B (Q4) gemma-2-27b · not orderable
- Gemma 2 2B (Q4) gemma-2-2b · not orderable
- Gemma 3 12B (QAT Q4) gemma-3-12b · preview
- Gemma 3 1B (QAT Q4) gemma-3-1b · preview
- Gemma 3 27B (QAT Q4) gemma-3-27b · preview
- Gemma 3n E4B (Q4) gemma-3n-e4b · preview
Stability AI Community Licence
Stability AI permits commercial use free of charge for organisations generating under US$1,000,000 in annual revenue, and requires the “Powered by Stability AI” attribution shown above. Full terms: Stability AI Community Licence Agreement. Stability’s Acceptable Use Policy applies to what you generate with them. You own the outputs, to the extent applicable law allows.
- SD-Turbo sd-turbo · preview
- SDXL-Turbo sdxl-turbo · preview
Apache License 2.0
Permissive. The notice-retention duty attaches to redistributing the weights; we do not redistribute them, so nothing further is required of you when you run one.
- AREX Turbo (Q4) arex-turbo-4b · preview
- Devstral Small (Q4) devstral-small · preview
- Devstral Small 2 24B (Q4) devstral-small-2-24b · preview
- Gemma 4 12B (QAT Q4) gemma-4-12b · preview
- Gemma 4 26B A4B (Q4) gemma-4-26b-a4b · preview
- Gemma 4 31B (Q4) gemma-4-31b · preview
- Gemma 4 E2B (Q4) gemma-4-e2b · preview
- Gemma 4 E4B (Q4) gemma-4-e4b · preview
- GPT-OSS 20B gpt-oss-20b · preview
- MiniCPM4 8B (Q4) minicpm4-8b · preview
- Mistral Small 3.1 24B (Q4) mistral-small-3.1-24b · preview
- Mistral Small 3.2 24B (Q4) mistral-small-3.2-24b · preview
- olmOCR-2 7B (Q4) olmocr-2-7b · preview
- Pixtral 12B pixtral-12b · preview
- Qwen2-VL 2B (Q4) qwen2-vl-2b · preview
- Qwen2.5 Coder 7B (Q4) qwen2.5-coder-7b · preview
- Qwen2.5-VL 32B (Q4) qwen2.5-vl-32b · preview
- Qwen2.5-VL 7B (Q4) qwen2.5-vl-7b · preview
- Qwen3 0.6B (Q4) qwen3-0.6b · preview
- Qwen3 14B (Q4) qwen3-14b · preview
- Qwen3 30B A3B (Q4) qwen3-30b-a3b · preview
- Qwen3 32B (Q4) qwen3-32b · preview
- Qwen3 4B (Q4) qwen3-4b · preview
- Qwen3 8B (Q4) qwen3-8b · preview
- Qwen3 Coder 30B A3B (Q4) qwen3-coder-30b · preview
- Qwen3-Next 80B A3B (Q4) qwen3-next-80b · preview
- Qwen3-VL 2B (Q4) qwen3-vl-2b · preview
- Qwen3-VL 30B A3B (Q4) qwen3-vl-30b-a3b · not orderable
- Qwen3-VL 4B (8-bit) qwen3-vl-4b · preview
- Qwen3-VL 8B (Q4) qwen3-vl-8b · preview
- Qwen3.5 4B (Q4) qwen3.5-4b · preview
- Qwen3.5 9B (Q4) qwen3.5-9b · preview
- Qwen3.6 27B (Q4) qwen3.6-27b · preview
- Qwen3.6 35B A3B (8-bit) qwen3.6-35b-a3b · preview
- Qwen3.6 35B A3B (Q4) qwen3.6-35b-a3b-4bit · preview
- Segmind Vega segmind-vega · preview
- SmolLM3 3B (Q4) smollm3-3b · preview
- SmolVLM2 2.2B smolvlm2-2.2b · preview
- SSD-1B ssd-1b · preview
MIT License
Permissive, same handling as Apache 2.0.
- DeepSeek-R1 Distill 14B (Q4) deepseek-r1-distill-14b · preview
- Fara 7B (Q4) fara-7b · preview
- Fara 1.5 9B (8-bit) fara1.5-9b · preview
- GLM-4.5 Air (Q4) glm-4.5-air · not orderable
- GLM-4.7 Flash (Q4) glm-4.7-flash · preview
- GLM-OCR (Q4) glm-ocr · preview
- GLM-OCR (8-bit) glm-ocr-8bit · preview
- MagenticBrain 8B (Q4) magentic-brain-8b · preview
- Phi-4-mini (Q4) phi-4-mini · preview
Vendor-specific licence
A licence written by the model’s publisher rather than a standard one. These stay unorderable until the specific terms have been read.
- LFM2-VL 1.6B (Q4) lfm2-vl-1.6b · not orderable
Research / non-commercial only
The publisher permits research and evaluation but not commercial use. Selling compute time on these would breach the licence, so they are permanently unorderable here. They are listed rather than hidden so it is clear the omission is deliberate.
- Nanonets-OCR2 3B (Q4) nanonets-ocr2-3b · not orderable
- Qwen2.5-VL 3B (Q4) qwen2.5-vl-3b · not orderable
Models without a recorded licence
A handful of older catalog entries predate this page and carry no licence field. They are listed here rather than quietly omitted; each is resolved as it comes up for release.
- Apple NLEmbedding (English, 300-dim) apple-nl-en-300d · live
- BGE-base (768-dim) bge-base · not orderable
- BGE-small (384-dim) bge-small · not orderable
- Blender Cycles (Metal) blender-cycles-metal · not orderable
- DeepSeek-R1 671B (Q4) deepseek-r1-671b · not orderable
- DeepSeek-R1 Distill 70B (Q4) deepseek-r1-distill-70b · not orderable
- DeepSeek V3 (Q2) deepseek-v3-q2 · not orderable
- E5-Mistral-7B e5-mistral · not orderable
- LLaVA 1.6 7B llava-1.6-7b · not orderable
- Mistral Nemo 12B mistral-nemo-12b · not orderable
- Mixtral 8x7B (Q4) mixtral-8x7b · not orderable
- Parakeet TDT 0.6B (Japanese) parakeet-tdt-0.6b-ja · not orderable
- Parakeet TDT 0.6B v2 parakeet-tdt-0.6b-v2 · not orderable
- Parakeet TDT 0.6B v3 parakeet-tdt-0.6b-v3 · not orderable
- Parakeet TDT-CTC 110M parakeet-tdt-ctc-110m · not orderable
- Phi-3 Medium 14B (Q4) phi-3-medium-14b · not orderable
- Phi-3.5-mini (Q4) phi-3.5-mini · not orderable
- PyTorch BYOM pytorch-byom · not orderable
- Qwen 2.5 14B (Q4) qwen-2.5-14b · not orderable
- Qwen 2.5 3B (Q4) qwen-2.5-3b · not orderable
- Qwen 2.5 72B (Q4) qwen-2.5-72b · not orderable
- Qwen 2.5 7B (Q4) qwen-2.5-7b · not orderable
- Qwen 2.5 Coder 14B (Q4) qwen-2.5-coder-14b · not orderable
- Qwen 2.5 Coder 32B (Q4) qwen-2.5-coder-32b · preview
- Qwen2-VL 7B qwen2-vl-7b · not orderable
- QwQ 32B (reasoning, Q4) qwq-32b · not orderable
- Apple Vision OCR vision-ocr-default · preview
- H.264 (VideoToolbox) vt-h264 · preview
- HEVC (VideoToolbox) vt-hevc · preview
- ProRes (VideoToolbox) vt-prores · preview
- Whisper Large v3 whisper-large-v3 · preview
- Whisper Small whisper-small · preview
- Whisper Tiny whisper-tiny · preview
- Xcode 15.x xcode-15 · not orderable
- Xcode 16.x xcode-16 · not orderable
Corrections
If you are a model publisher and something on this page misstates your licence, or a model of yours is listed that you would rather not see on a compute marketplace, email legal@commoncompute.ai and we will correct or remove it. We would rather be told than be right.