Skip to main content

Every measured API, on one page

Each figure was measured on the production service against a public or stated test set. Higher is better throughout. Every row links to the API page that explains it and to the paper with the dataset, method and errors.

31 APIs measured 21 papers

Documents

The business-document and receipt figures come from synthetic test sets with invented values; real documents score lower, as their paper states.

Identity and face

Speech

Language and translation

  • 75.8%Thai Sentiment Analysisposts labelled correctly as positive, neutral or negative2,614 Thai social-media postsOct 2026PDF
  • 82.4%Thai Question Answeringtoken F1: how far each answer matches the reference answer739 Thai Wikipedia questionsOct 2026PDF
  • chrF++ 56.1Multilingual TranslationThai to English: closeness to a professional translation, 0 to 1001,012 sentences in 13 directions (FLORES-200)Oct 2026PDF
  • chrF++ 54.6ChindaMT TranslationThai to English: closeness to a professional translation, 0 to 1001,012 sentences, both directions (FLORES-200)Oct 2026PDF

Large language models

  • 83.1%OpenThai 2.0Thai national exam questions answered correctly443 exam questionsOct 2026PDF
  • 0.772OpenThai 2.0 Legalcitation F1: how far the cited law sections match the correct ones, 0 to 1200 legal questions (NitiBench-CCL)Oct 2026PDF

Images

  • 196/200iApp Image Upscalerimages closer to the original than standard resizing, at 4x200 public reference imagesOct 2026PDF

Vehicles

All research