Where the HSK data comes from.

Vocabulary, character and dictionary data

Counting method

Source rows are rows in the input lists before HSK Atlas normalization: 5,000 current-HSK rows and 11,000 HSK 3.0 vocabulary rows.

Normalized level assignments are the final version-and-level memberships after variant splitting and duplicate removal: 5,000 current-HSK assignments and 10,968 HSK 3.0 assignments.

Unified vocabulary entries merge the same simplified headword across datasets while preserving all memberships. The current build contains 11,061 entries. One headword may still represent more than one reading or dictionary sense.

Counting method 2; normalization 1; generator 1.

Known limits

• Dictionary senses can be broader than the meaning intended in a syllabus row.
• For a small, explicit set of high-ambiguity beginner words, HSK Atlas places the common learner meaning before broader CC-CEDICT senses; the override map is stored in the data generator.
• A unified simplified headword is not necessarily one independent word sense.
• Some notation entries cannot be mapped safely to a dictionary gloss and remain under review.
• The HSK 3.0 advanced vocabulary is published as one combined Levels 7–9 range.
• Exam rollout can change; candidates should verify the exact format shown for their booking.

Last data build

July 27, 2026 UTC. Editorial status verified 2026-07-23.