Skip to main content
The Urantia Papers are available in Spanish: every paragraph of the Foreword and the 196 papers, and every paper, section, and part title. The Spanish is an AI translation, made with Claude Opus 5.5 in October 2026, revised in November 2026, and released as CC0 (public domain). No human translator wrote it. Each translated paragraph has a label that says so.

Get the Spanish text

Add lang=es to a request. Without lang, the API returns English, as before.
These endpoints accept lang: With include, the Papers parallels (urantiaParallels) come with their Spanish text and titles, and the entity mentions (entities) with their Spanish names. The Bible and the other works in an include stay in English.

Search in Spanish

Keyword search with lang=es searches the Spanish text. A word matches with or without its accents (“religion” finds “religión”), and “documento 1” is a reference, like “1:0.1”. The results come as an exact phrase first, then paragraphs with every word. Spanish has no typo fixes.
Semantic search with lang=es takes a question in Spanish or English and returns Spanish paragraphs. A question in Spanish scores lower than the same question in English, so strength and meta.related allow for that difference.

A translated paragraph

The other fields (id, htmlText, labels, audio, and the rest) are the same as in English. htmlText holds the Spanish text with the same markup. Show the label to your readers, for example “AI translation”. Do not present the Spanish as an official or human translation.

In the MCP server

The paper, paragraph, and search tools take lang: "es": toc.get, papers.list, papers.get, papers.sections, paragraphs.get, paragraphs.context, paragraphs.random, search.fulltext, and search.semantic. Each Spanish paragraph has language and translation, as in the REST API. See MCP Servers.

In the SDK

@urantia/api 0.7.0 and later for the papers and paragraphs, 0.8.0 and later for search:

The works in Spanish

The Bible and the other works are English texts, and their text stays English. Their catalog words are in Spanish: the titles, traditions, summaries, unit and division labels, chapter titles, and the names of the Bible books. These are an AI translation too, released as CC0. Each work has textLanguage: "eng", in every language. With lang=es it also has a catalogTranslation label, the same shape as translation on a paragraph. A word that has no Spanish yet stays English. The MCP tools works.list, works.read, and bible.books take lang: "es" too, and so do works.list, works.read, and bible.books in @urantia/api 0.9.0.

What stays in English

  • Audio. All narration is English.
  • Topic labels on papers and paragraphs.
  • Some entity descriptions. Spanish has a description for 2,517 entities; the others stay in English. See Entities.
  • The text of the Bible and the other works. Their titles and labels are in Spanish: see The works in Spanish.

How the translation was made

Why a new translation

The published Spanish translations of the Papers are under copyright. An app, a website, or an AI assistant cannot copy them or serve them without permission. This translation is CC0: anyone can copy it, change it, and use it, with no permission and no credit.

How it was made

  1. We built our own glossary, from the English. The Papers use about 2,560 special terms: beings, places, orders, and words with a meaning of their own, such as “morontia” and “personality”. For each term, an AI model found its meaning in the English text and chose one Spanish word. For each key concept and each term with less than high confidence, a second model chose a word on its own, without the first choice. Another model saw only the Spanish words and said what each one suggests. Where these results disagreed, a separate model decided. The models were Claude Opus 5.5, Claude Sonnet 5.5, and Claude Fable 5.1. No published translation was a source.
  2. Claude Opus 5.5 translated each paper, in a new session for each paper, with three things only: the English text of the paper, our glossary, and our style rules. The model never saw a published Spanish translation, and it could not search the web.
  3. A program checked each paragraph for copying. It is a program, not an AI model. It compares our text with the published Spanish editions and gives only a number: how much of the wording matches. Two human translations share about 13 to 34 percent of their wording, because both follow the same English. A paragraph above 60 percent went back to a new session, which got the English again and none of the old wording. The check is there because an AI model can remember a published translation from its training.

How good it is

In November 2026 we tested the translation with a blind review. A reviewer read the English paragraph and several Spanish versions side by side, with no names, so it did not know which version was ours. For each version it gave a score from 1 to 5 for accuracy (the same meaning as the English) and for fluency (natural Spanish), and it listed each error as major (a change of meaning) or minor. It also ranked the versions from best to worst.
  • Four versions: ours and the three published Spanish editions: 1993 (Latin America), 2009 (Europe), and 2021 (International).
  • The sample: 80 paragraphs from 10 papers (5, 12, 27, 41, 59, 85, 99, 120, 143, and 174), from all four parts of the book.
  • Eight reviewers from three AI companies: Anthropic (Claude Opus 5.5 two times, Claude Sonnet 5.5 two times, and Claude Fable 5.2), OpenAI (GPT-6 Astra and GPT-6.1 Sol), and Google (Gemini 4 Carbon). Each reviewer worked on its own and read all 80 paragraphs, so the test has 640 reviews.
Our version ranked above the 2021 edition in 81 percent of the reviews, above the 2009 edition in 82 percent, and above the 1993 edition in 95 percent. The result is the same when the reviewers are not from Anthropic, the company that made the translating model: The main difference is accuracy. In fluency, our text and the 2021 edition are close (4.63 and 4.55).

What the test does not show

  • The reviewers are AI models. A review by people who speak Spanish is a stronger test, and we have not done one yet.
  • 80 paragraphs are a sample, not the whole book.

How the fixes were made

After each review, we read the notes of the reviewers on our text. We fixed a note only when we agreed with it. Separate AI models made the fixes, and they saw only three things: the English, our Spanish, and the problem, written in English. They never saw a published translation or a Spanish wording from a reviewer. Each fix then had to pass the copy check again. The check compares the paragraph with each published edition. A fix failed when the paragraph matched an edition more than 60 percent, or when the fix made the match more than 10 points higher. A failed fix went back to a new session and was written again from the English. A fix must also keep our glossary and our term decisions. A reviewer does not know them, so a note can ask for a word that we chose on purpose not to use. For example, our text uses “adoración” for “worship” and “veneración” for “adoration”, to keep the two English words apart. The fixes:
  • October 2026, after an earlier review with fewer versions and reviewers: the names of peoples are lowercase, as Spanish writes them (“los noditas”, “los anditas”), in 264 paragraphs. “Orden” is masculine for an order of beings, in 144 paragraphs. Small fixes in 19 other paragraphs.
  • November 2026, after the test above: 103 paragraphs.
    • 25 errors of meaning, grammar, and wording that the reviewers found.
    • The same problems in other paragraphs of the book. “Revolve” became “orbitar” where the English means an orbit, in 16 paragraphs. “Self-conscious” became “autoconsciente”, in 23 paragraphs. “Should” became “debería” where the English gives advice beside a “must”, in 30 paragraphs.
    • 9 paragraphs that now follow the term decisions.
  • November 2026, after a read of every flagged paragraph: 32 paragraphs.
    • The automatic checks flagged about 1,240 paragraphs. Most flags were false alarms: an English word in its ordinary sense (“actual” for “real”, “love” as a verb), or a correct Spanish form that the check did not see. We made the checks better, and we read each remaining flag with its paragraph and the glossary.
    • 30 paragraphs got fixes. “Devil” became “diablo” in 11 paragraphs. The technical “actual” became “en acto” in 5. Other fixes are “indwelling”, “star cloud”, “pattern”, “ghost worship”, “adánico” for consistency, and 2 words that a sentence had dropped.
    • 2 paragraphs read too close to a published edition, and we wrote them again from the English.
Each paragraph also has an automatic check for its length, its numbers, and its terms. A paragraph that a check flags has translation.review: "needs_review". A flag that we read and found wrong is recorded for that exact text, so a later change to the paragraph is checked again.

Style

The text uses “ustedes” for the plural “you”, lowercase pronouns for God (“él”, “su”), and « » for quotations, so that it reads well in every Spanish-speaking country.

Limits

  • It is an AI translation, and no human translator wrote it. A sentence can read too close to the English, and a word can have the wrong sense.
  • It is not a translation of Urantia Foundation, and the Foundation did not review or approve it.
To report a wrong word or an error, use Feedback and give the reference of the paragraph. A fix in the API reaches every app that uses it.