AlignED: Benchmarking AI Models for Educational Practice

Abstract

Teachers are rapidly adopting generative AI. Data from the OECD's Teaching and Learning International Survey (TALIS), collected in 2024, shows that 41% of lower secondary teachers across participating countries have used AI for professional tasks. Yet few evaluations test models on the knowledge and judgements that matter in educational settings: understanding how learning works, reasoning about why teaching approaches succeed or fail, demonstrating the pedagogical knowledge expected of qualified teachers, and comparing student work against curriculum standards.

AlignED addresses this gap with five evaluations spanning neuromyth identification, diagnostic reasoning, teacher certification knowledge (general pedagogy and inclusive education), and student work judgement. A total of 32 models from five providers have been tested across five evaluations. Key findings: no single model ranks first across all evaluations; performance on one benchmark does not predict performance on another; assessment task type matters enormously, with models performing well on comparative judgement tasks but achieving only "fair" chance-adjusted agreement (Cohen's κ = 0.25) on standards-based grading; and no model expressed uncertainty on any confidence probe, even when answering incorrectly.

This report is a frozen snapshot of the February 2026 benchmark data. For the full list of AlignED reports, visit AlignED Reports.