Where the map comes from

The map follows the official HSK 3.0 syllabus: 11,000 current HSK 3.0 words, plus 147 legacy HSK 2.0 cross-reference entries. Pinyin and part of speech are read from the official list files. The older HSK 2.0 levels are kept alongside as a second track, so if your textbook or exam date still runs on the old lists, you can study those too.

HearMandarin is an independent study aid. We are not affiliated with, or endorsed by, the organizers of the HSK exam.

Meanings start from the CC-CEDICT dictionary (Creative Commons BY-SA 4.0) and are turned into short teaching glosses in three tiers. The counts below come straight from the site's own ledger at build time:

• 1,920 words carry a gloss a human editor has reviewed.
• 9,227 words carry a gloss prepared by machine from the dictionary and cleaned for learners.
• 0 words are waiting for review and are shown straight from the dictionary in the meantime.

Spot a gloss that reads oddly? Tell us and we fix it.

Every example sentence, graded reader, and survival scene is written in-house, then machine-checked against the official word lists for its level. If a page says Band 2, every word on it is Band 2. That check is a gate, not a vibe, which is why the readers carry a machine-checked badge.

The 220 readers each carry comprehension questions with explanations, and every explanation points back to the line in the story that proves it.

The 720 practice questions are written in-house and run through the same machine check against the level's word list, so a Band 1 question stays inside Band 1. Every question carries a short explanation of the answer.

The word list itself is open data. The HSK 3.0 words with pinyin, band, part of speech and our teaching glosses are published as CSV and JSON under a Creative Commons Attribution 4.0 license, so you can build your own flashcards, apps or research on top of it. Credit HearMandarin with a link and you are done.

Open the dataset page for the field list, the download links and the current version date.

The audio here is one steady voice per role, produced rather than recorded in a room, which is why word one and word eleven thousand sound like the same person on the same day. Every clip goes through a check that compares what was said with the text it should say; whatever the check flags gets listened to, and clips that fail are made again. Same-sounding words and tone-only slips can hide from that check, so if a clip still sounds wrong to your ear, tell us and we fix it. Slow playback is a built in player feature on the site; the offline files are normal speed.

Official lists get revised, and we re-read them when they do; every check is dated on the HSK 3.0 status page. Spotted something wrong? Tell support@hearmandarin.com and we will fix it and say so.