Kimi K2.6, the best-performing open-source model, achieves just 3.8% on Diamond, 16% on Main and 37% on Extended.

if you can effectively posttrain a model to only meaningfully perform with your closed source agent, then you get to funnel the majority of users to your agent at the expense of your model/API co-opetition

DeepSeek-V4-Pro-Max beats GPT-5.2 and Gemini 3.0-Pro on standard reasoning benchmarks and lands just behind GPT-5.4 and Gemini 3.1-Pro

While our production codebase has significantly diverged, including major rewrites of core systems like authentication and data handling, we want to ensure there is still a truly open version available.

focusing on the ~1.5K mainline open models from the likes of Alibaba's Qwen, DeepSeek, Meta's Llama

focusing on the ~1.5K mainline open models from the likes of Alibaba's Qwen, DeepSeek, Meta's Llama

Donations

We’re now relaunching PRO, but instead of a paid chat and (never existing) paid documentation, your team gets access to paid gems, our visual editor for workflows, and a commercial license.

Better community building: At the moment, MDN content edits are published instantly, and then reverted if they are not suitable. This is really bad for community relations. With a PR model, we can review edits and provide feedback, actually having conversations with contributors, building relationships with them, and helping them learn.

Choose Your Plan