Aristotle says its AI tutor beats frontier tutoring modes in its own pedagogy test

Nov 20, 2026
In the startup's evaluation of 791 tutoring moments, its system passed 54% against 33% to 35% for prompted OpenAI, Google and Anthropic models.

AI tutoring startup Aristotle says its tutor gives students the right amount of help more often than the tutoring modes of OpenAI, Google and Anthropic models, in an evaluation the company designed and ran itself and published on its blog.

Aristotle sampled 500 live sessions from its own platform and isolated 791 "pivotal turns" in guided problem-solving: a student's first attempt, a correct step, a wrong answer, or getting stuck. LLM judges, calibrated against labels from experienced human tutors, scored each response on one question: did the tutor provide the right amount of support? On that measure, Aristotle passed 54% of turns, against 34% for GPT-6.1 Sol on OpenAI's Study Mode prompt, 33% for Gemini 3.8 Flash on Google's Guided Learning prompt and 35% for Claude Sonnet 5.5 on the same Guided Learning prompt.

The company says frontier models are post-trained to solve problems from start to finish, so they tend to set up equations or reveal the next step before the student has tried, and prompting alone does not fix that. It also reports a median time to first word of 1.6 seconds for its multi-agent system.

The limits are in Aristotle's own notes: the rival systems ran as reconstructions of OpenAI's and Google's tutoring prompts inside Aristotle's replay harness rather than the consumer apps, the turns come from Aristotle's own sessions, and the scores measure next-turn decisions, not learning outcomes. The company calls 54% "just the starting point."

We hope you enjoyed this article

Consider subscribing to one of our newsletters like AI in Education or Daily AI Brief.

Also, consider following us on social media:

Free newsletter

AI in Education

Weekly newsletter about AI in education. Covers AI-driven software for educators, schools, general innovations and regulatory updates.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.