Back to Blog
Product Update • Sep 1, 2026 • 3 min read
Faster, Cheaper Document Embeddings
We moved our document indexing to a faster embedding model — documents are ready to chat about in seconds, not minutes.
A chatbot is only useful once your documents are indexed. Long upload times were the #1 complaint we heard from new users, so we rebuilt the indexing path.
What changed
- New embedding model — we now use a faster text-embedding model that produces high-quality chunks in a fraction of the time.
- Parallel chunking — document splitting and embedding now run concurrently across chunks.
- Up to 5x faster indexing on typical PDFs and DOCX files.
The result
Most documents are now ready to chat about in under 10 seconds. Large 100+ page files that previously took a minute now index in well under 30 seconds.
| Document | Before | After |
|---|---|---|
| 10-page PDF | ~10s | ~2s |
| 50-page PDF | ~40s | ~9s |
| 150-page PDF | ~90s | ~25s |
No action is needed — existing chatbots use the new indexer automatically the next time you upload or re-index.
Artem K.
Creator of DocChat. Indie creator and product person.