Codex Sassoon

An almost complete Hebrew Bible written about the year 900, sold at Sotheby's New York on 17 May 2023 for a hammer price of USD 33.5 million and USD 38.1 million with fees, the highest sum ever paid at auction for a manuscript. It carries its own ownership record in ink: a sale note of about the year 1000 naming the men who bought and sold it, a dedication to the synagogue at Makisin in what is now north-eastern Syria, and a later clause entrusting it to a named guardian against the day the synagogue would be rebuilt. It never was.

Then the trail stops for roughly five centuries. The codex reappears in Turkey in 1929, is bought for 350 pounds, passes through a British pension fund that treated art as an inflation hedge, through an unnamed dealer, and through a Geneva financier, before a ninety-four-year-old former American ambassador placed an irrevocable bid on it and gave the money to a museum in Tel Aviv. Every page of it can be downloaded free, in colour, at full resolution.

It is 792 pages, and the losses are argued about

The codex is a single bound volume of 396 parchment leaves, 792 pages, written three columns to the page in iron-gall ink on what the Times of Israel describes as sheepskin. It measures roughly 30 by 36 centimetres. It contains all twenty-four books of the Hebrew Bible, with Tiberian vowel points, cantillation marks, and the Masoretic apparatus in the margins.

How much is missing is not agreed. Sotheby's material as reproduced by Medievalists.net says eight leaves are gone; Fine Books and Collections says twelve full leaves; the Associated Press puts survival at about 92 per cent of the text; one account counts fifteen missing chapters, ten of them in Genesis. The metadata attached to the colour scan, which is derived from the object itself rather than from a press release, records that Genesis begins partway through chapter 11. The losses cluster in Genesis and in Psalms, and hundreds of further leaves are partly damaged rather than absent.

The binding is not original and not even second. Sotheby's suggests the book was rebound for the first time at Makisin in the thirteenth century. The Sotheby's specialist Shaul Seidler-Feller places another rebinding at the end of the Middle Ages or the beginning of the modern period, probably by non-Jewish binders. What holds it today is a plain twentieth-century leather binding. The bookplate of David Solomon Sassoon is pasted inside it.

The apparatus is the work of two men, and the second erased the first

Yosef Ofer of Bar-Ilan University, presenting at the research workshop on the codex held jointly by Bar-Ilan and ANU in January 2025, set out a finding that changes how the manuscript should be read. The Masorah in the codex is the work of two Masoretes working at different times. Through the first ninety pages the second man added his comments alongside those of his predecessor. Later in the volume he stopped adding and began erasing, scraping out the earlier notes to write his own in their place.

The two are distinguishable by what interests them. The first rarely deals with plene and defective spelling. The second deals with it constantly, and he worked with the Aleppo Codex and with the Babylonian Masorah in front of him. Neither is named anywhere in the book.

That erasure is a physical fact about the object with a consequence for the market. A buyer is not acquiring one act of copying but a layered document in which a later editor destroyed part of an earlier one. Nothing in the sale coverage said so.

The scribe was careless, and scholars say so plainly

The reverence in the auction coverage is not shared by the people who read the manuscript for a living. Israel Yeivin, in his standard introduction to the Tiberian Masorah, records that the scribe frequently omits punctuation, diacritical marks and vowels, and errs in consonantal spelling on dozens of occasions. Where a vowel is repeated on consecutive consonants he often marks only the first. Alephs habitually receive an ordinary shva instead of a hataf vowel. The sof passuq is sometimes forgotten at the end of a verse.

Kim Phillips of Cambridge University Library, writing before the sale, put it in one word to the Associated Press: the work is surprisingly sloppy next to the Aleppo Codex, and a scholar choosing between them would take Aleppo without hesitation.

Yeivin measured the affiliation as well. Where the traditions disagree, this codex follows Ben Asher about 40 per cent of the time, Ben Naphtali about 20 per cent, and neither about 40 per cent. It is not a clean witness to a single school. That is precisely what makes it useful: Vince Beiler, mapping some 45,000 Masorah parva notes across 110 early codices, places it neither at the centre of the surviving Masoretic Bibles nor at their periphery, and reads that in-between position as consistent with a tenth-century date and a Syrian origin.

The book records its own sale, in about the year 1000

Most medieval manuscripts have to have a provenance reconstructed for them. This one wrote part of its own down.

An early note attests the sale of the codex by Khalaf ben Abraham, described in one account as a businessman active in Palestine and Syria, to Isaac ben Ezekiel al-Attar, who later transferred ownership to his two sons, Ezekiel and Maimon. Fine Books and Collections dates the note to the early eleventh century. It is worth being exact about the names, because at least one published account renders the second son as Maimonides, which he was not.

No source consulted reproduces the wording of that note in the original. Every English version of it circulating in 2023 is a paraphrase of the Sotheby's catalogue. The scan is public, and the note is in it; a transcription and translation from the manuscript itself is the obvious next piece of work on this record.

Makisin, and a condition that was never met

A second inscription, added by an anonymous hand, consecrates the book to the Lord God of Israel and gives it to the synagogue of Makisin. Makisin is the medieval name of the place now called Markada, in the Al-Hasakah governorate of north-eastern Syria.

The town was destroyed. By whom is not settled: the destruction is attributed either to the Mongols later in the thirteenth century or to the Timurids at the start of the fifteenth, and the second reading fits better with the date of the note that follows.

That note, written on the last page and dated by Fine Books and Collections to the late fourteenth century or later, is the one that matters most to anyone reading this as a title document. The codex was entrusted to Salama ibn Abi al-Fakhr on condition that he return it to the synagogue once it was rebuilt. If the rebuilding did not happen in his lifetime, his sons would inherit the book on the same terms.

The synagogue was never rebuilt. The inheritance clause is the detail almost every account drops, and it is the clause that legitimately moved the codex out of communal ownership and into a family. Everything that follows, including a price of 38.1 million dollars, runs through it.

Then nothing, for about five centuries

After the Salama note the record stops. Seidler-Feller, writing as Sotheby's own specialist, says the trail goes cold at the end of the Middle Ages or the beginning of the modern period, at the point where the book was rebound, and that after hundreds of years of silence it suddenly reappears in Turkey.

The gap is usually reported as six hundred years. It is closer to four hundred and fifty or five hundred, from the late fifteenth century to 1929, and the looser figure should not be repeated. It is a genuine hole in the chain and no reconstruction has been offered for it.

1929: a Frankfurt librarian who could not afford it

The reappearance has been reported two incompatible ways. The Associated Press wire says the codex was acquired in Frankfurt in 1929. Wikipedia says it was bought from an owner living in what is now Ankara. Both are partly right and the workshop paper reconciles them.

Seidler-Feller records correspondence of March and April 1929 between Aron Freimann of the Stadtbibliothek in Frankfurt am Main and David Solomon Sassoon. The manuscript had surfaced in Turkey. Freimann could not afford the owner's asking price, so he brought it to the attention of his friend Sassoon, who then bought it for considerably less than had first been asked.

The price was 350 pounds. Sassoon gave it the number 1053 in his own collection and published it in his library catalogue, Ohel Dawid, in 1932, which is how it entered scholarship under the name it carried until 2023. Photographs of it reached Jerusalem in the late 1950s. It has since been renumbered again: it is now ANU MS 1.

1978: the pension fund that bought art

In 1978 the Sassoon family sold the codex, together with thirty-two other manuscripts from the collection, at Sotheby's in Zurich. The buyer was the British Rail Pension Fund, and the reported price is about 320,000 dollars. No source consulted states whether that figure is the hammer or the total with fees.

The fund is the point of interest, not the price. In 1974 the trustees of the British Rail pension scheme decided to put a small part of the fund into works of art as a hedge against inflation, eventually committing something in the region of 40 million pounds, about 3 per cent of holdings, across purchases made between 1974 and 1981. It was the first and, so far as is recorded, the only large British pension fund to do it. Sotheby's advised on the buying without charging a fee, on the condition that any eventual sales went through Sotheby's. The fund began selling in 1986 and finished in 2003.

So the manuscript was, for eleven years, an investment asset held against the retirement of British railway workers, and the auction house that advised on the purchase had contracted in advance for the right to sell it. Reported annual returns on the programme, in the region of 13 per cent against roughly 22 per cent on the fund's equities over the same span, come from secondary reporting and should be checked against the academic literature before they are relied on.

1989: two sales in one year, and only one of them is reported

Eleven years later the fund put the codex back into the market at Sotheby's in London, on 5 December 1989. The result was 2,035,000 pounds.

The buyer was not Jacqui Safra. Seidler-Feller states that the successful bidder was a Jewish antiquarian book dealer, unnamed, who promptly sold it on to Jacob, called Jacqui, Eli Safra of Geneva in the same year. The figure of 3.19 million dollars that appears in almost every account of the 2023 sale is the dollar conversion of the London auction result, not the price Safra paid. What Safra paid the dealer has never been disclosed.

Whether 2,035,000 pounds was a hammer figure or a total including the buyer's premium is not stated in any source consulted, and this record does not guess. One outlier is in print: ReligionUnplugged gives Safra's price as 4.19 million dollars against everyone else's 3.19 million, which looks like a transposition but has not been corrected.

Safra kept it for thirty-four years. Over that period the codex was exhibited publicly once, at the British Museum in 1982, which was during the pension fund's ownership rather than his.

17 May 2023: the lot was effectively pre-sold

Sotheby's offered the codex in New York as a single-lot sale titled Codex Sassoon: The Earliest Most Complete Hebrew Bible, lot 1, catalogued as Circa 900, Land of Israel or Syria, estimated at 30 to 50 million dollars.

The gavel fell at 33.5 million dollars. With fees the total was 38.1 million. Both figures are stated in the reporting and this record keeps them apart, because a great deal of coverage gave only the second and called it a price.

The structure behind those numbers is the part almost nobody printed. Alfred H. Moses had placed an irrevocable bid of 32.5 million dollars with Sotheby's before the sale opened. An irrevocable bid means the lot cannot fail: the bidder is contractually committed at that level and the auctioneer may bid up to, but not at, that amount on the book. The codex was therefore already sold at 32.5 million before anyone in the room raised a hand. When an anonymous rival reached 33 million, Moses went one increment further to 33.5 and it was over. The Art Newspaper describes a four-minute duel between two buyers; the wider lot, with a handful of bidders in the room and on the telephone, took under six minutes.

The underbidder has never been named. The underbid is known, which is more than most sales disclose, and at 33 million dollars it establishes that the price had real depth behind it rather than resting on one determined buyer.

One thing does not reconcile, and it is not a rounding error. The gap between 33.5 million and 38.1 million is about 4.6 million dollars, an effective buyer's premium of 13.7 per cent. Premium scales are tiered downwards, in bands that fall as the price rises, which means the effective rate on a lot of this size can never drop below the top band. No major house charged as little as 13.7 per cent at the top of its scale in 2023. So either the gavel price was lower than 33.5 million or the total was higher than 38.1 million, and the two numbers that every account of this sale prints side by side cannot both be true. This record prints both and resolves neither.

Who bought it, and in what capacity

Alfred H. Moses is a lawyer, a former United States ambassador to Romania and a former president of the American Jewish Committee. He was ninety-four at the time of the sale, and he had never seen the book. He first looked at it about two weeks afterwards.

The capacity in which he bought it is reported inconsistently and the distinction matters for a provenance record. The Art Newspaper describes him as acquiring it on behalf of the American Friends of ANU. Sotheby's own specialist describes the buyer as ANU, courtesy of Moses, who donated the money required for the purchase. Moses himself put it as deciding to buy it and provide the funds for ANU to buy it, and the irrevocable bid stood in his own name.

The reading this record adopts is that Moses funded the purchase and the buyer of record was the American Friends of ANU. He gave the money, not the object.

His stated reason is on the record and is worth quoting because it is a statement about what the price was for: he wanted it somewhere it would be available to the Jewish people rather than in a rich person's bank vault.

Whether it is the earliest most complete Hebrew Bible

The sale title is a claim, and it was contested before the hammer fell.

Three manuscripts frame the question. The Aleppo Codex, of about the same date, was the most authoritative text in the tradition and lost roughly 40 per cent of its leaves, including nearly all of the Pentateuch, in the twentieth century. The Leningrad Codex, of 1008 or 1009, is entirely complete but more than a century younger. This codex sits between them: older than Leningrad, far more complete than Aleppo. It is neither the oldest nor the most complete, and the claim rests on the combination.

Kim Phillips of Cambridge University Library attacked the wording directly, calling claims of the form the earliest most complete Hebrew Bible nonsensical, on the ground that a text can be complete or incomplete but not most complete, and grossly misleading. He noted that another codex, the Aleppo Codex, has a competing claim to be the earliest Masoretic Hebrew Bible containing the entire text in a single volume, and that being the earliest surviving example does not make something the first ever made.

There is a harder objection than the wording. The second Masorete of this codex saw the Aleppo Codex and copied at least two Masorah magna notes from it, as Neria Klein of the Hebrew University has established, and a marginal note refers to Aleppo by its Arabic name a-Taj, the Crown. That constrains the date of the second man's layer to after Aleppo. It does not date the original copying, and it should not be reported as if it did, but it is the substantive reason the earliest claim is disputed rather than merely loose.

The case for the manuscript is made in the same scholarly venue. Noam Mizrahi of the Hebrew University calls the Masoretic codex a conceptual revolution and this its earliest sample to date. Beiler argues that its value lies partly in its difference from Aleppo and Leningrad rather than its proximity to them.

What the record actually is, and what it is not

Sotheby's, the Art Newspaper and Fine Books and Collections all fenced the claim to manuscripts. On that fencing 38.1 million dollars broke a record that had stood since November 1994, when the Codex Leicester made 30,802,500 dollars including fees at Christie's New York.

It is not the most expensive item of text ever auctioned. In November 2021 a 1787 printing of the United States Constitution made 43.2 million dollars with fees at Sotheby's, against the underbidding of ConstitutionDAO, a crowdfunded group that raised roughly 47 million dollars in Ether from some 17,000 contributors and could not bid past about 40 million and still afford conservation. That lot is classed as a printed document rather than a handwritten manuscript, and the distinction is doing real work in every headline written about both sales.

Delayed by a war, then put on permanent display

Before the sale the codex toured: Sotheby's London from 22 to 28 February 2023, Tel Aviv in the week of 20 March, Dallas in the week of 17 April, Los Angeles in the week of 30 April, and New York in early May. The Art Newspaper reports more than 20,000 visitors across the tour.

It flew to Israel on El Al from New York on 5 October 2023 and was to go on permanent display on 11 October. The attacks of 7 October intervened and the display was held. It finally opened on 29 May 2025, ahead of Shavuot, some twenty months late, opened at the page carrying the Ten Commandments in Deuteronomy 5.

It is now in the gallery called The Foundations on the first floor of ANU in Tel Aviv. The museum describes the codex as entrusted to it for safekeeping on behalf of the Jewish people and humanity, which is a formulation about custody rather than ownership and is worth noticing in a document about title. No source dated 2026 was found; the present tense here rests on the permanence of the 2025 installation, not on a fresh report.

The whole book is free, at full resolution

This is the part a student should know and almost nobody writes down. The entire codex was photographed in colour by Ardon Bar-Hama for the National Library of Israel and was made accessible through the library's Ktiv portal two years before the sale.

The complete colour set is on the Internet Archive, uploaded in February 2023, before the auction, at 9.2 gigabytes, under a Public Domain Mark with any use permitted. There are JP2 archives of 5.2 and 3.0 gigabytes and a 344 megabyte PDF. A separate black and white microfilm scan is there as well. Wikimedia Commons carries the full colour scan and twenty-four further files, one per book of the Bible, plus a single page image at 7,099 by 8,344 pixels.

No IIIF manifest specific to the codex was located from the National Library of Israel, ANU or Sotheby's, and the absence is recorded here as unverified rather than established. What is established is that anybody can read every surviving page of a manuscript that cost 38.1 million dollars without asking permission or paying anything, which is not true of most objects in this archive.