humans& announces Ampersand Tutor for its agent API

humans& announced Ampersand Tutor (&Tutor) for its agent API and says it trained against Persimmon, a learner model. It reports TutorMoments overall scores of 0.754 versus its starting model’s 0.500 under a plain prompt, and 0.919 versus GPT-6 Astra’s 0.926 under an evaluation-aware prompt. humans& says the benchmark draws on real tutoring sessions but scores responses using an AI student and an LLM judge, not human learning outcomes.