Open Thoughts
Open source reasoning datasets projectCurated datasets for reasoning models
Highlights
What you can do with it
- Training a reasoning model on curated data
- Developing new reasoning algorithms with the tools
- Using the datasets as teaching material
How to use it
- 1 Read the project background on the site
- 2 Browse the datasets and performance results
- 3 Download the datasets and Evalchemy
- 4 Train your model and evaluate it with Evalchemy
- 5 Follow the blog for updates
Who it is for
Pros
- Free access to curated datasets
- Built with multiple research institutions
- Includes an evaluation toolchain
Limitations
- Aimed at technical users with training experience
- Dataset scope is limited to reasoning tasks
- Free access and project direction can change
What Open Thoughts is
Open Thoughts is a community project led by Bespoke Labs and the DataComp community. Its goal is to curate high quality open source reasoning datasets for training advanced small models, because dataset quality is often the limiting factor in maths and code reasoning. It brings together researchers and engineers from institutions including Stanford, UC Berkeley, and the University of Washington. The project supports benchmark testing for maths and code reasoning, ships the Evalchemy tool for evaluating models, publishes the latest performance results for the community, and shares progress through a blog. Access is currently free, and the material is aimed at people who actually train or study reasoning models.
What you can do with it
- Download curated open source reasoning datasets
- Benchmark maths and code reasoning with the provided setups
- Evaluate your own models with Evalchemy
- Read published performance results to compare approaches
- Follow the blog for updates and new releases
Who it is for
- Researchers training or studying reasoning models
- Developers building new reasoning algorithms
- Educators using real datasets to teach model training
What to watch out for
This is a technical resource, not a product you install and use: you need experience with model training and evaluation to get anything out of it. The focus is reasoning, especially maths and code, so it will not cover general purpose data. Access is free now, but the datasets, tools, and benchmarks can be revised or replaced as the project moves, so pin versions rather than relying on a snapshot. Always cite the source and check the licence before you build on the data, and expect to spend real compute on training rather than treating the data as a plug and play solution.
FAQ
What is Open Thoughts?
A community project curating open source reasoning datasets for training small models.
Who is behind it?
Bespoke Labs and the DataComp community, with contributors from universities including Stanford, Berkeley, and Washington.
Is it free?
Yes, it is currently free to access.
Why it is free
What the official page says： 免费
Free limits, trial length and features are decided by the official site and can change at any time.
Product preview
Screenshot archived from the listing page, for preview only.
