Tic Tac Go LLM Benchmark

July 13, 2026

This is a page that gets updated like a wiki. It represents my thoughts at the particular time I last edited it. The thoughts expressed here may not be the thoughts I have now.

This benchmarks how efficiently an LLM can solve the Google Search engine game https://share.google/fictsJUJSq36as1hQ. They can use any tools they want

7/13/2026

used cursor harness with screenshot and https://github.com/lazy-sock/tic-tac-go provided

Composer 2.5 non-fast: 2min 15s, $0.14
Grok 4.5 High: 5min 6s, $0.05
GPT 5.6
- Luna
  - none: Fail
  - low: 1min 27s, $0.17
- Terra
  - none: 49s, $0.38