Rewind.ai for Open Source
Pro access at no cost for active open-source contributors.
Every Self-Hosted Model Is Open-Source
Every self-hosted model on Rewind.ai is open-source under permissive licenses (Apache 2.0, MIT, BSD). No Llama/Meta licenses. We think AI should be transparent, auditable and free to use.
Chat & Language Models
| Model | Use Case | License | Source |
|---|---|---|---|
| Qwen2.5-72B | General-purpose chat, reasoning and analysis | Apache 2.0 | HuggingFace |
| Qwen2.5-7B | Lightweight chat for low-latency tasks | Apache 2.0 | HuggingFace |
| Mistral-7B | Compact general-purpose chat | Apache 2.0 | HuggingFace |
| Phi-3 | Compact reasoning and math tasks | MIT | HuggingFace |
Code Models
| Model | Use Case | License | Source |
|---|---|---|---|
| Qwen2.5-Coder-32B | Code generation and debugging | Apache 2.0 | HuggingFace |
| DeepSeek-Coder-V2-Lite | Code completion and review | MIT | HuggingFace |
Image & Video Models
| Model | Use Case | License | Source |
|---|---|---|---|
| FLUX.1-schnell | Image generation with low latency | Apache 2.0 | HuggingFace |
| Stable Diffusion XL | Detailed image generation | OpenRAIL++ | HuggingFace |
| Kandinsky 2.2 | Stylized and artistic image generation | Apache 2.0 | HuggingFace |
| CogVideoX-2B | Text-to-video generation | Apache 2.0 | HuggingFace |
| AnimateDiff | Animating still images into short clips | Apache 2.0 | HuggingFace |
Speech & Audio Models
| Model | Use Case | License | Source |
|---|---|---|---|
| Kokoro | Text-to-speech synthesis | Apache 2.0 | HuggingFace |
| Piper | Fast on-device text-to-speech | MIT | GitHub |
| MeloTTS | Multilingual text-to-speech | MIT | GitHub |
| Chatterbox | Expressive text-to-speech | MIT | HuggingFace |
| OuteTTS | Lightweight text-to-speech | Apache 2.0 | HuggingFace |
| faster-whisper | Speech-to-text transcription | MIT | GitHub |
| Vosk | Offline speech-to-text | Apache 2.0 | GitHub |
| AudioLDM 2 | Audio and music generation | Apache 2.0 | HuggingFace |
| Demucs | Stem separation from mixed audio | MIT | GitHub |
Translation, Vision & Other Models
| Model | Use Case | License | Source |
|---|---|---|---|
| MadLAD-400 3B | Translation across 450+ languages | Apache 2.0 | HuggingFace |
| Real-ESRGAN | AI-based image upscaling | BSD | GitHub |
| BRIA RMBG 2.0 | Automated background removal | Apache 2.0 | HuggingFace |
| SAM | Segment any region in an image | Apache 2.0 | GitHub |
| DeOldify | Colorize black-and-white photos | MIT | GitHub |
| all-MiniLM-L6-v2 | Sentence and text embeddings | Apache 2.0 | HuggingFace |
| Tesseract | Optical character recognition | Apache 2.0 | GitHub |
Open Source Maintainer Program
Maintain a public open-source project? We give active contributors free Pro access. Send us a linkto your repository and we'll get you set up.
FAQ
Qwen2.5-72B handles complex reasoning and multi-step analysis well. Qwen2.5-7B and Mistral-7B respond faster and cover most everyday tasks. Phi-3 is a compact choice when you need reliable reasoning or math without waiting long.
Yes. If you built or know of a model under a permissive license, reach out. We review submissions for output quality, license terms and real-world usefulness before adding anything to the platform.
If you actively maintain a public open-source project, you can apply for free Pro access. Send us a link to your project and we will review it. Qualifying maintainers get their account upgraded at no cost.
Before a model goes live, we check benchmark results, actual output quality across common tasks, inference latency and license compliance. Models that do not clear those bars are not added, regardless of popularity.
FLUX.2 covers image-to-image tasks including upscaling, background removal and photo colorization. Recraft handles high-detail enhancement work.