#1: baba HebrewCore
Best for: Production apps, learning products and text-to-speech pipelines that need nikud with a safety check and a published contract.
- POST /v1/nikud: full vowel points, up to 10,000 characters per request; English terms and numbers left untouched.
- verified: true when stripping the points returns your text; when false, your text comes back unchanged.
- POST /v1/strip-nikud: removes nikud, dagesh and cantillation, not counted against the quota.
- Published plans (Free 100,000 characters a month, Starter $49), quota headers and a status page at hc-status.itsbaba.com.
- add_nikud and strip_nikud are tools on the HebrewCore MCP server.
#2: Dicta Nakdan
Best for: Free nikud for research and non-commercial work.
Where it beats baba HebrewCore: Free to use, where HebrewCore nikud counts against a character quota (3x weight).
From Dicta's pages, as of September 2026
- Dicta presents Nakdan tools on its site for adding nikud (vocalization) to Hebrew text. Source
- A third-party client describes the Nakdan API as free with no formal published contract, and its endpoints as version-numbered. No Dicta page stating API terms, pricing or an SLA was found. Source
#3: Dicta open models (self-hosted)
Best for: Teams that want to host nikud themselves with no per-character fee.
Where it beats baba HebrewCore: No license fee and no per-character cost once you run it yourself, which HebrewCore cannot match on price. You operate the model, the servers and the updates.
From Dicta's pages, as of September 2026
- dictabert-large-char-menaked, a nikud restoration model, is published on Hugging Face under CC-BY-4.0. Source
- These are models you download and host, not a hosted API. Source