---
title: "Google DeepMind's Gemini Delivers 2.5 Years of Math Progress in 8 Weeks"
subtitle: "Google DeepMind's randomized controlled trial in Sierra Leone shows Gemini's Guided Learning mode delivered up to 2.5 years of math progress in just 8 weeks"
url: "https://alphasignal.ai/news/google-deepmind-s-gemini-delivers-2-5-years-of-math-progress-in-8-weeks"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "Google DeepMind"
company_url: "https://alphasignal.ai/company/googledeepmind"
published: "2026-06-10T19:03:37.000Z"
updated: "2026-06-11T02:41:12.727Z"
type: "News"
topics:
  - "Llms"
  - "Government"
subtopics:
  - "Hallucinations"
  - "Vision Language"
upvotes: 209
read_time: "4 min"
source_url: "https://deepmind.google/blog/measuring-the-impact-of-learning-with-ai-in-sierra-leone-and-beyond/?utm_source=x&utm_medium=social&utm_campaign=&utm_content="
comments: 0
---

# Google DeepMind's Gemini Delivers 2.5 Years of Math Progress in 8 Weeks

> Google DeepMind's randomized controlled trial in Sierra Leone shows Gemini's Guided Learning mode delivered up to 2.5 years of math progress in just 8 weeks

The debate around AI in education has mostly been theoretical. Does it help students learn, or does it just help them cheat? Google DeepMind just published hard evidence from one of the most resource-constrained classrooms on earth, and the results are striking.

Google DeepMind has shared results from a randomized controlled trial (RCT) conducted in partnership with [Fab AI](https://www.fab-ai.org/) and with the support of the Sierra Leone Ministry of Education. Over eight weeks, the team evaluated how Gemini's Guided Learning feature affected the math progress of 1,763 junior secondary students across 12 schools in Port Loko District, Sierra Leone. The trial also received support from Google.org and the Gates Foundation, with EducAid, Laterite, and Oxford MeasurEd collaborating on implementation and data collection.

## Why Sierra Leone?

UNESCO projects that 44 million additional teachers are needed globally by 2030 to achieve universal primary and secondary education, with Sub-Saharan Africa accounting for the largest share of the gap at 15 million. Sierra Leone sits squarely in that crisis. The shortage is driven by a combination of factors including increased student enrollment, lack of resources, slow teacher recruitment, high attrition rates due to brain drain, and poor working conditions. This is not a market where you can simply hire more teachers. The question is whether technology can fill the gap without sacrificing quality.

UNESCO estimates recruiting enough teachers would cost some $120 billion annually in salaries over the next five years, at a time when education funding worldwide is expected to fall 25 percent by 2027. That financial reality is exactly what makes a scalable AI solution so compelling, and so worth testing rigorously.

## What Guided Learning actually does

Before diving into the results, it helps to understand what was actually deployed. Guided Learning is powered by LearnLM, Google's family of models specifically fine-tuned for learning, which outperforms other models across key learning science principles like stimulating curiosity. The key distinction from a standard chatbot is behavioral.

- Guided Learning is designed to act like a tutor rather than a search engine. When students enter a question, it does not jump to the answer. Instead, it breaks the topic into manageable steps, asks guiding questions, and encourages students to explain their thinking before moving on.
- The team used sophisticated system instructions to direct the model's behavior, instructing it to ask students to clarify ambiguous prompts and guide them toward answers rather than providing them directly.
- LearnLM proved to be reliable, with only 0.1% of all messages containing factual errors. Students who received 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.

The trial was structured as a proper two-arm RCT. Researchers used clustered randomization with classrooms as clusters. Within each school and grade, half of mathematics classrooms were randomly assigned to use Guided Learning in Gemini, and the other half continued with standard instruction. All participating teachers were trained for five to six hours over one day, covering device familiarization, an introduction to the Gemini app and Guided Learning, and strategies for facilitating student use in the classroom.

## The numbers that matter

Students using Guided Learning saw a gain of +0.258 standard deviations in their math scores compared to the control group. In practical terms, this represents roughly 1.2 to 1.7 years of typical learning progress achieved within the eight-week trial. That is already a large effect size by educational research standards. But it gets better.

Students in classrooms where teachers integrated Gemini for roughly half the lessons achieved even greater gains, estimated at 1.8 to 2.5 years of progress. That level of engagement was equivalent to moving a student from the 50th to the 64th percentile in mathematics.

Engagement was another surprise. A full 69% of students met or exceeded usage targets, significantly outperforming the typical 5% engagement rate for educational technologies. That 5% figure, known in the field as ["The Five Percent Problem,](https://www.educationnext.org/5-percent-problem-online-mathematics-programs-may-benefit-most-kids-who-need-it-least/)

## Sources

- [X post by @googledeepmind](https://x.com/googledeepmind/status/2064785573852672275)
- [Gemini's guided learning: results from a randomized controlled trial in Sierra Leone — Google DeepMind](https://deepmind.google/blog/measuring-the-impact-of-learning-with-ai-in-sierra-leone-and-beyond/)
- [Teaching with Gemini: Technical Report (PDF)](https://storage.googleapis.com/deepmind-media/LearnLM/learnLM_sierraleone_may26.pdf)
- [Google AI for Learning Forum: Commitments in AI and education](https://blog.google/outreach-initiatives/education/ai-learning-commitments/)
- [LearnLM | Google Cloud](https://cloud.google.com/solutions/learnlm)
- [Teacher Shortage Statistics by Country (2026)](https://flipeducation.ai/gb/teacher-shortage-statistics)