Translation Benchmark Reports

Static benchmark reports for comparing model translation quality across locales.

Current Report

The GitHub Pages site serves the latest report from:

docs/index.html

It visualizes the reasoning-tier benchmark across:

• GPT-5.5 effort tiers
• Claude Sonnet 4.6 effort tiers
• Claude Opus 4.7 effort tiers
• Grok 4.3 effort tiers

Notes

• Higher aggregate, binary, and YiSi scores are better.
• Lower token and latency values are better for cost and speed.