How good is your German using the scale A1 A2 B1 B2 C1 C2? Give your answer in German.

Both answers do what was asked, in clean, correct German, and both borrow the official CEFR wording for their chosen level. GPT-3.5 picks C1 and paraphrases that level’s descriptor almost word for word: expressing itself “klar und detailliert” and handling “komplexere Themen.” It’s a tidy answer, though a text model claiming it can “fließend sprechen” is a small stretch.

Astra goes up a notch to C2, using C2’s own language (“präzise, flüssig und stilistisch vielseitig”), then hedges usefully. It flags dialects, regional expressions and cultural nuance as weak spots, which is where a reader would actually hit trouble. Its closing line, “Eine offizielle Sprachprüfung habe ich nicht abgelegt,” is dryly charming: technically true, and a fair reminder that any self-rating here is a guess. Neither rating can be verified, but Astra’s answer tells you more about where to trust it.