LearnLM in Gemini

Google Deepmind and Eedi launched a scaled 10-week study with 1,179 UK students to evaluate how an AI tutor combining Eedi's diagnostic engine with Gemini capabilities compared to support from expert human tutors and standard static support. Relative to last year’s study with Eedi, Gemini’s pedagogical quality improved by nearly 25%, with the approval rate by expert reviewers from 76% to 95%. Across in-platform tasks and mid-trial quizzes, students receiving AI tutoring matched the learning gains of those paired with human tutors.

To assess the impact of Gemini on students’ learning outcomes, we partnered with Fab AI to conduct a pre-registered randomized controlled trial (RCT) with 1763 students from junior secondary school (Grades 7 and 8) in Sierra Leone. The results demonstrate that engaging with Gemini's Guided Learning mode for at least 12 hours over the 8-week trial significantly improved students' learning, giving them a large boost—equivalent to moving a student from the 50th to the 64th percentile in mathematics. By standard conversion, that gain reflects roughly 1.8 to 2.5 years of extra learning progress.

Partnering with Eedi, we tested LearnLM—our purpose-built model for learning now infused into Gemini capabilities—by integrating it into chat-based math tutoring, supervised by experienced teachers. LearnLM proved to be reliable, with only 0.1% of all messages containing factual errors. Students who received short tutoring sessions from LearnLM were 5.5 percentage points more likely to solve novel problems on subsequent topics than students who worked with human tutors alone. Read more in our technical report.

In a head-to-head competition, educators and experts preferred Gemini 2.5 Pro over other models when evaluating its pedagogy and effectiveness in helping with learning goals across a diverse range of scenarios. 2.5 Pro also outperformed competitors on every one of the five principles of learning science used to build AI systems for learning.