Model Golf
Build the best language model under 100 million parameters. Minimum 1028-token context. Winner receives $50 in RunPod credits. Open source only. Monthly rounds.
What It Is
✓What Model Golf Is
- ✓ A monthly parameter-efficiency competition
- ✓ Open to anyone, anywhere
- ✓ Fully custom architectures allowed
- ✓ Novel tokenization & training schemes welcome
- ✓ Open source by design
- ✓ $50 RunPod credits to the winner
✗What It Is Not
- ✗ Not a Kaggle-style dataset contest
- ✗ Not restricted to existing architectures
- ✗ Not limited by llama.cpp compatibility
- ✗ Not a fixed-benchmark leaderboard
- ✗ Not for closed-source submissions
The Rules
Simple and clear. Complexity discourages participation.
How To Join
The first competition launches soon. Here is how it will work.
Build
Create a language model under 100M parameters with at least 1028-token context. Any architecture. Any approach.
Document
Open source your repository with full documentation. Include parameter count, context window, and approach summary.
Submit
Submit your repo link when the competition opens. Submissions are evaluated monthly. Winner gets $50 in RunPod credits.
Why Small Models?
Parameter efficiency is the path to democratized AI.
Small models run on consumer hardware. They deploy at the edge. Individuals can fine-tune them on a single GPU. Model Golf rewards architectural creativity within tight constraints, making efficiency competitive.
The $50 prize is symbolic — enough to cover GPU time, enough to motivate enthusiasts, not enough to attract mercenaries. Enthusiasts build interesting things. Interesting things advance the field.
Ready To Compete?
The first Model Golf competition launches soon. Join the Discord for updates.