Benchmarks

Original tests of AI models. Each one runs every model once on a fixed question set, checks the answers against sourced keys, and publishes what the run really cost.

Multilingual LLM Bench

Which AI speaks your language best. 150 questions written from scratch in 30 languages, 17 models, run 2026-10-02. Leader: GPT-6 Astra at 70.7.

Knowledge Cutoff Bench

17 AI models tested on 132 dated questions, 4 a month from Jan 2024 to Sept 2026. Each model's real cutoff next to the cutoff its lab claims. Run 2026-10-15. Leader: GPT-6 Sol at 25.0.

Canadian Tax LLM Bench

11 AI models scored on 130 questions about the 2025 Canadian tax year, 10 for each province and territory, with every answer computed from official rate tables. Run 2026-10-15. Leader: GPT-6 Astra at 80.8.