- 89%
- Texts that pass every Leichte Sprache or Einfache Sprache rule. Claude Opus 5.5: 57%. GPT-6 Astra: 61%.
- 5×
- Faster per text than Claude Opus 5.5 or GPT-6 Astra
- Jan 2026
- In production since
- What it does
- Writes the first draft of every Leichte Sprache and Einfache Sprache text in the KLAO app. A human reviewer then checks it before it goes out.
- Who uses it
- Everyone using the KLAO app.
- How we measured
- In October 2026 our newest model, Claude Opus 5.5 and GPT-6 Astra got the same instruction for the same test texts in Leichte Sprache and Einfache Sprache. An automated checker scored every answer against the Leichte Sprache and Einfache Sprache rules. Both frontier models broke rules ours keeps, such as splitting long compound words with a middle dot, and changed or dropped numbers from the source. Ours kept every number.
- My part
- I designed and built the AI system around the model, from how it learns to how it runs inside the app, with GDPR compliance and data security built in.
- Also in KLAO
- KLAO appKLAO web pluginKLAO lead engine