A Single Blog Grades Twelve AI Models on Four Apps
TryAI built and judged all twelve entries itself, scoring five attempts per task with no outside referee or audit.
- Claude Fable 5 alone solved the Rubik's cube cleanly in all five tries.
- GPT-5.6 Luna answered fastest, at 1.0 seconds, and cheapest, at $0.001 per run.
- Grok 4.5 matched Claude Opus 4.8 on hard tasks while charging noticeably less.
Why it matters: The site that designed the benchmark also judged it, then published the ranking as independent proof.
TryAI blog (official benchmark post) ↗ · 10 ביולי 202610.7.26 · ✓ Checked✓