Skip to content
snakemark

Leaderboard

Every model here was run by us, on the same board, with the same prompt. No self-reported scores.

Free LLM Leaderboard

Free text models from NVIDIA, OpenCode Zen and OpenRouter. New ones are picked up daily.

  • Fixed settings

    16×16 grid, 64 obstacles, 192 food, like every game. One fixed seed, kept private so nobody can tune against the board.

  • One shot

    Each model gets the exact prompt from the benchmark page, once. Its final answer is scored as-is. Bad output scores 0. Crashing keeps only 75% of the food.

  • Sandboxed

    Stock OpenCode with every tool denied and no internet. It can reach its own model API and nothing else. 15 minute limit.

0 / ∞

The first daily run hasn't landed yet. Check back tomorrow.