Open Source AI Models
30 open-source and open-weight AI models ranked by quality. These models can be self-hosted, fine-tuned, and deployed without vendor lock-in.
30
Open Source Models
21
Language Models
10
Providers
8
Free to Use
Open Source Language Models
| # | Model | Quality | API Price |
|---|---|---|---|
| 1 | Qwen3 235B A22B Alibaba | 87.0 | $0.46 / $1.82 |
| 2 | Qwen3 Max Alibaba | 85.0 | $0.78 / $3.90 |
| 3 | DeepSeek R1 DeepSeek | 85.0 | $0.70 / $2.50 |
| 4 | R1 0528 DeepSeek | 83.0 | $0.50 / $2.15 |
| 5 | Qwen2.5 Coder 32B Instruct Alibaba | 82.0 | $0.66 / $1.00 |
| 6 | Command A Cohere | 80.0 | $2.50 / $10.00 |
| 7 | Command R+ (08-2024) Cohere | 79.0 | $2.50 / $10.00 |
| 8 | Llama 4 Scout Meta | 79.0 | $0.08 / $0.30 |
| 9 | DeepSeek V3.2 DeepSeek | 77.0 | $0.20 / $0.77 |
| 10 | Llama 3.1 70B Instruct Meta | 77.0 | $0.40 / $0.40 |
| 11 | DeepSeek V3 DeepSeek | 76.0 | $0.23 / $0.91 |
| 12 | Llama 4 Maverick Meta | 76.0 | $0.15 / $0.60 |
| 13 | Phi 4 Microsoft | 74.0 | $0.07 / $0.14 |
| 14 | Command R (08-2024) Cohere | 73.0 | $0.15 / $0.60 |
| 15 | Mistral Large Mistral | 73.0 | $2.00 / $6.00 |
| 16 | Mistral Nemo Mistral | 72.0 | $0.02 / $0.03 |
| 17 | Mistral Small 3.1 24B Mistral | 72.0 | $0.35 / $0.56 |
| 18 | Llama 3.3 70B Instruct Meta | 71.0 | $0.10 / $0.32 |
| 19 | Qwen2.5 72B Instruct Alibaba | 71.0 | $0.36 / $0.40 |
| 20 | Llama 3.1 8B Instruct Meta | 68.0 | $0.02 / $0.05 |
| 21 | Command R7B (12-2024) Cohere | 65.0 | $0.04 / $0.15 |
Other Open Source Models
Whisper Large v3
OpenAI · Speech-to-Text
FLUX.1 Dev
Black Forest Labs · Image Generation
Stable Diffusion 3.5
Stability AI · Image Generation
Wan 2.1
Alibaba · Video Generation
FLUX.1 Schnell
Black Forest Labs · Image Generation
Stable Audio 2
Stability AI · Music & Sound
CogVideoX 1.5
Zhipu AI · Video Generation
Stable Diffusion XL
Stability AI · Image Generation
MusicGen Large
Meta · Music & Sound
What does "Open Source" mean for AI?
Open-source AI models publish their model weights, allowing anyone to download, run, fine-tune, and deploy them. Some models (like Meta's Llama) use custom licences with usage restrictions, while others (like Mistral and some DeepSeek models) use fully permissive licences. "Open weight" is technically more accurate for models that share weights but not training data or code, though "open source" is commonly used.