Open Source AI Model Leaderboard — Best Open Weights (2026)
A ranked leaderboard of the best open-source and open-weight AI models across code generation, video generation, reasoning, and multimodal categories. Compare licenses, parameter counts, context windows, and benchmark scores to find the right open model for your deployment.
Rapidly Evolving Field: The specialized and emerging AI model landscape is advancing quickly. New models, benchmarks, and pricing changes occur monthly. Last reviewed: July 2026.
Open Source Model Rankings
9 open-source models ranked by category and capability. All models listed have publicly available weights under permissive licenses (MIT, Apache 2.0, or community licenses).
| Rank | Model | Provider | Category | License | Parameters | Context |
|---|---|---|---|---|---|---|
| 1 | DeepSeek Coder V3 | DeepSeek AI | Code & Developer AI Models | MIT License | Mixture-of-Experts (MoE) Decoder Transformer | 128,000 tokens |
| 2 | Qwen2.5-Coder | Alibaba Cloud | Code & Developer AI Models | Apache 2.0 License | Decoder-only Transformer (Qwen2.5 family) | 128,000 tokens |
| 3 | StarCoder2 | BigCode / HuggingFace | Code & Developer AI Models | BigCode Open Model License | Decoder-only Transformer (StarCoder family) | 16,384 tokens |
| 4 | Wan2.1 | Alibaba Cloud | Video Generation Models | Apache License 2.0 | Diffusion Transformer (DiT) | — |
| 5 | DeepSeek R1 / R2 | DeepSeek AI | Reasoning & Thinking Models | MIT (fully open, commercial use allowed) | Mixture-of-Experts (MoE) Transformer | 128,000 tokens |
| 6 | QwQ-32B | Alibaba Qwen Team | Reasoning & Thinking Models | Apache 2.0 (commercial use allowed) | Dense Decoder-only Transformer | 32,768 tokens |
| 7 | Llama 4 Maverick | Meta AI | Multimodal Frontier Models | Llama 4 Community License | Mixture-of-Experts (128 experts, 17B active) | 1,000,000 tokens |
| 8 | Kimi-VL | Moonshot AI | Multimodal Frontier Models | MIT | Dense transformer with vision encoder (early-fusion) | 128,000 tokens |
| 9 | InternVL3 | Shanghai AI Lab | Multimodal Frontier Models | MIT | Dense transformer with InternViT vision encoder (early-fusion) | 128,000 tokens |
Deploy Open Source Models
Open-source models can be self-hosted on your own GPU infrastructure or deployed via managed inference providers. Check each model's detailed page for hardware requirements and deployment guides.
Browse All Models →