The bindings are published on npm as tokenizers-wasm; the project says release candidates use the next dist-tag, with installation via npm install tokenizers-wasm@next.
The release also adds Python support for encoding pairs of texts. Its fixes cover byte-level BPE vocabularies missing bytes that cannot occur in UTF-8, including a cited ModernBERT case; WordPiece vocabularies containing an empty-string token; and tokenizer.json files using regex Split pre-tokenizers or Replace normalizers, which had been rejected by Python, Node, and WASM bindings.
Other changes make a 64-bit vocabulary-hash collision an error rather than silently dropping a token, remove an unnecessary Rust/Python encoding copy, and document feature selection through the umbrella Rust crate. The release notes say feature parity with version 0.23 remains the main milestone before stable 1.0.0. They do not establish a stable-release date, browser performance figures, or compatibility guarantees beyond the listed fixes.