Unlocking Collective Intelligence in AI Research
Tiny Labs is a community-driven research lab on a mission to build the world's first sub-billion parameter language model to surpass 80% accuracy on the MMLU benchmark. This performance is currently only achieved by models with trillions of parameters.
Real-time inference on any device
Run on phone, laptop, edge devices
1000x less energy consumption
Fraction of the cost to train and deploy
By focusing on small, efficient language models, we're creating a more accessible and sustainable AI ecosystem that empowers individuals and organizations to build sophisticated AI applications without the massive computational resources typically required.
Each month we release a new version of our model, accompanied by a detailed blog post explaining what changed and our progress. Our models improve continuously as we iterate on architecture, training data, and optimization techniques. Each benchmark represents our latest model's performance.
Inspired by EleutherAI and Marin, we're building a community-led approach to developing small language models. Anyone with a great idea can test it, refine it with peers, and get credit for their impact. Our mission is to make contributing to frontier AI as straightforward as contributing to open-source software.
Submit your innovative ideas and detailed proposals via our structured GitHub issues. Focus on architectural changes, optimiser tweaks, or data variations. No idea is too small!
On a bi-weekly/monthly cycle, the community and sponsors vote on the most promising proposals through an open auction system.
Selected experiments are evaluated against general benchmarks like MMLU, and potentially sponsor-defined tests, ensuring robust assessment.
Receive financial rewards, build a transparent reputation, and gain co-authorship on our monthly technical reports. Top contributions are merged into our main model. Small targeted changes (e.g., a new pre-processing step, quality filter, or hyper-parameter tweak) are fairly credited alongside larger architectural overhauls.
See what we've been up to.
Discover why Tiny Labs believes the next leap in AI is about people, not just parameters. Learn about our mission, vision, and the story behind our community-driven approach to building efficient, accessible language models.
Explore how our platform lowers barriers, tracks experiments, and rewards contributors. See how the experiment hub, job launcher, and community auction system enable open, collaborative AI research at scale.
Learn how to propose, implement, and get recognized for your contributions to Tiny Labs. This guide covers our step-by-step process, evaluation, rewards, and the values that drive our open research community.
We are Pietro and Richard, two Cambridge PhD grads. During our PhD, we shared an office; now we share the same passion for building the world's best small language models.
Co-Founder
Co-Founder