Moxin Organization

Introducing Moxin 7B: The truly open, SOTA-performing LLM and VLM that's redefining transparency.

We've open-sourced EVERYTHING—pre-training code, data, and models, including our GRPO-enhanced Reasoning model. It outperforms Mistral, Qwen, and LLaMA in zero-shot/few-shot tasks and delivers superior reasoning on complex math benchmarks, all with an efficient training cost of ~$160K for full pretraining.

We unleash the power of reproducible AI 🚀. Interested? Explore the models and code on our GitHub and read the full paper on arXiv.