← Back to Tech Radars

Context Golf

Craft the most efficient prompt to solve coding challenges. Pick your files, write your system prompt, define your tools — then see if a tiny in-browser model can nail it. Fewer tokens + smaller model + better output = higher score.

AI-generated content — verify before acting on any recommendations.
Challenges 7 holes
Select a challenge
Pick a challenge from the left panel to begin.
System Prompt 0 tokens
Context Files 0 tokens Click to toggle, right-click or long-press to preview
Load a challenge to see available files.
Tools / Skills 0 tokens
Examples:
Your Prompt 0 tokens
Total: 0 tokens
Model Output
Run a challenge to see the model's response here.
Scorecard
Model Settings
Token Budget 0 tokens
System Files Tools Prompt
Best Scores
Complete a challenge to see your scores here.
How Scoring Works
Final Score = Quality × Efficiency × Model Bonus × 100
Validation
Required patterns must all match. Failure = DNF and no score recorded.
Quality (30–100%)
Line-by-line similarity to the reference answer. Any submission that passes validation earns at least 30% — but a concise, targeted answer matching the expected output line-for-line scores much higher.
Efficiency
Par ÷ tokens used. Under par > 1.0, so fewer tokens = better multiplier.
Model Bonus
Smaller models multiply your score. SmolLM2 135M = 2.5×, Phi 3.5 Mini = 0.5×.
Golf Rating
Token-only: Ace (≤50%), Eagle (≤75%), Birdie (under par), Par (at par), Bogey (≤125%), Double Bogey (over 125%).

A verbose essay can pass validation but score near-zero on quality. Aim for the specific insight the challenge asks for, not a comprehensive audit.

Ready — pick a challenge
Simulation mode