AGIEval
AGIEval is a human-centric evaluation benchmark built from 20 official, public admission and qualification exams intended for real human test-takers. It was assembled by Ruixiang Cui with collaborators from Microsoft and released alongside the 2023 paper "AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models". The current release, version 1.1, contains 8,062 items across 20 tasks (18 multiple-choice and 2 cloze), spanning English and Chinese. Sources include the American SAT, US law school admission tests (LSAT), graduate and management admission material in the style of GRE and GMAT, mathematics competitions (AMC and AIME), and the Chinese College Entrance Exam (Gaokao) across subjects such as English, history, biology, chemistry, physics and mathematics, plus Chinese civil service and lawyer qualification questions. Access is straightforward and ungated: the items ship as JSON (with passage, question, options, label and answer fields) in the repository under data/v1_1, cloned directly from git. For RAG work it offers clean, structured, level-graded practice items with known answers, which suits retrieval-augmented question answering and evaluation harnesses. For an AI tutor its value is authentic exam-prep material: real admissions and qualification questions that can seed practice sets, calibrate difficulty, sequence learners toward specific tests and support mastery modelling against verified answer keys, in two languages. On licensing, the repository code and annotations carry an MIT licence, so attribution is required and there is no share-alike obligation. The exam content itself is a different matter: the maintainers state that all use of the data should follow the licence of the original datasets, and the underlying exams (SAT, LSAT, GMAT, GRE, Gaokao, AMC/AIME) are copyrighted, so per-exam reuse must be confirmed before any commercial redistribution. Unlike curriculum sources we list such as openstax, khan-academy, ck-12 and siyavula, or standards like common-core and ngss, AGIEval is assessment and benchmark material rather than teaching content; it complements maths practice collections such as amps and stackmathqa by adding graded, exam-authentic test items.
Related sources
Achievement Standards Network
Machine-readable curriculum standards from US states, national bodies and other jurisdictions, modelled as an RDF graph of URI-addressable learning objectives with cross-jurisdiction alignments. Now run by D2L and free to use.
CASE Network (1EdTech)
A public registry of machine-readable learning-standard frameworks from all 50 US states and other issuing agencies, run by 1EdTech in the CASE JSON format. The digitally referenceable spine of what to teach, at which level and in what order, rather than the teaching content itself.
CEFR Companion Volume Descriptors
The Council of Europe's 2020 CEFR Companion Volume descriptor set, the de facto international standard for levelling language proficiency from Pre-A1 to C2. A spine of can-do statements for sequencing and grading language teaching.
CEFR-J English Profiles (Open Language Profiles)
Open datasets from Tono Lab at TUFS mapping English vocabulary and grammar to fine-grained CEFR-J sub-levels, from A1 to C2. Grade words and structures by level and gauge text difficulty for language tutoring.