OpenAI Model Craft launches Parameter Golf challenge
Parameter Golf is OpenAI’s open research challenge to build the most efficient pretrained model under extreme constraints. Participants try to minimize held-out loss on a fixed FineWeb dataset while staying within a 16 MB artifact limit for weights plus training code and a 10-minute training budget on 8×H100s.
This is less a product launch than a high-signal research contest with a recruiting funnel attached.
- –The constraint set is the story: 16 MB total and 10 minutes of training forces efficiency work, not brute force scaling.
- –The GitHub PR workflow makes it feel like an engineering competition with a public leaderboard, not a marketing landing page.
- –OpenAI is clearly using this to surface talent and potentially publicize winning methods, which gives the challenge real incentive beyond bragging rights.
- –The compute-credit offer lowers the barrier enough to attract serious entrants without turning it into an unrestricted bake-off.
DISCOVERED
45d ago
2026-04-16
PUBLISHED
74d ago
2026-03-18
RELEVANCE
AUTHOR
OpenAI