Oumi Summer ’26 is here.
See what we launched
Oumi Open Weights Directory
Choose a model you can own.
Model leaderboards are a starting point. The decision is your task, your output contract, your hardware, and your right to deploy. This directory turns that into a short list you can test.
Model profiles
Alibaba Cloud · China
What is Qwen?
Qwen is Alibaba Cloud’s open-weight model family. It spans small models that are practical to fine-tune through larger mixture-of-experts releases.
Active in Oumi →
DeepSeek · China
What is DeepSeek?
DeepSeek publishes open-weight language models, including large mixture-of-experts systems. The family is often shortlisted where reasoning and coding matter.
Listed by Oumi →
Meta · United States
What is Llama?
Llama is Meta’s open-weight model family. Its wide ecosystem makes it a familiar baseline for teams evaluating or fine-tuning their first custom model.
Listed by Oumi →
Mistral AI · France
What is Mistral?
Mistral is a European model family spanning open-weight and commercial releases. The right starting point depends on the exact release and deployment constraints.
Listed by Oumi →
Google · United States
What is Gemma?
Gemma is Google’s open-weight model family. It includes compact checkpoints that can be useful when hardware limits and deployment footprint drive the decision.
Listed by Oumi →
Comparisons
Best model guides
Best Open-Source LLM
There is no universal winner. Pick a model family, then evaluate the exact checkpoint against your real task, output contract, deployment budget, and license constraints.
Best LLM for Coding
Use a coding evaluation that resembles your real work: the languages, repositories, tools, response format, latency target, and cost ceiling. Then test a smaller specialist before renting a large general model forever.
Oumi benchmark data is coming.
We’ll add reproducible task benchmarks here once the evaluation recipe, model version, dataset, hardware, and scoring criteria are published. Until then, this directory doesn’t invent a winner.