In his book Die Babylonier. Geschichte, Gesellschaft, Kultur (2004; 3rd, revised edition 2015), Michael Jursa (Professor of Assyriology at the University of Vienna)
writes that the following categorisation of cuneiforms texts has established itself:
Archival texts are texts that belong to the sphere of daily life. They include contracts, letters, administrative documents/notes, settlements of accounts, etcetera.
They constitute the majority of cuneiform texts (tablets and fragments):
Grob geschätzt fallen 80% der schriftlichen Überlieferung Mesopotamiens in diese Kategorie.
I.e. roughly 80% of Mesopotamian texts that have come down to use fit into this category.
These texts constitute a unique source of information about the social and economic history of Mesopotamia, since no other civilisation from the same period (except for Egyptian papyri from late Antiquity) has left us a similar corpus.
- Monumental texts are mainly royal inscriptions on buildings and objects, and other texts intended for public display. These can sometimes include useful historical information, but they tend to glorify the ruler who ordered the construction or creation of the object on which the inscription can be found. Jursa does not give an estimation of the share of these texts in the overall corpus.
- Canonical texts are those texts that were studied in by scribes in "schools" (these were not public institutions like modern schools) and passed on from generation to generation.
The category of texts can be subdivided into scientific texts and literary texts. "Scientific texts" are those that were used by certain professions, such as scribes, evocators, diviners and doctors.
(In the chapter on "Babylonian science", Jursa also discusses lexical lists used for scribal training.)
In the chapter on Babylonian literature, Jursa explains that modern readers may be more interested in the literary texts, but this does not change the fact that it is the more practical texts that constitute the core of the "canon".
Jursa does not give an estimation of the share of canonical texts in the overall corpus.
Michael Jursa and Juan Carlos Moreno García also discussed this classification in Chapter 4 "The ancient Near East and Egypt" in Fiscal Regimes and the Political Economy of Premodern States (edited by Andrew Monson and Walter Scheidel, Cambridge University Press, 2015).
The categorisation appears to have been proposed by William W. Hallo, who used also used it in a three-volume work he co-edited: The Context of Scripture: Canonical Compositions, Monumental Inscriptions and Archival Documents from the Biblical World.
However, in Cuneiform Texts and the Writing of History, Marc Van de Mieroop points out that
We have no simple criterion to differentiate between literary and non-literary texts, as style is a difficult concept to use in a textual tradition so alien to ours, and as the Mesopotamians themselves did not indicate a generic categorization of their writings which would help us.
So the percentage of cuneiform texts (clay tablets or fragments) that can be categorised as literature is definitely below 20%, but it is not clear from the above sources how much below 20%.
In the article "Großes Fach Altorientalistik: Der Umfang des keilschriftlichen Textkorpus" (Mitteilungen der deutschen Orientgesellschaft zu Berlin, 2010),
Michael P. Streck published estimations of the size of the text corpus of Ancient Near East texts that have come down to us.
This obviously does not cover all texts that were ever written down in the Ancient Near East until 300 A.D. (since much has been lost), nor texts that have not been discovered yet.
The estimations are based on word counts, not the number of tablets or tablet fragments. Chapter 3.10 of his article discusses "canonical texts" (see Hallo's categorisation, above).
For literary texts, he offers an estimation of around 100.000 words. (For comparison, he adds in a footnote the word counts for Iliad and the Odyssey, namely 115.477 and 87.765 words, respectively.)
After the discussing the share of "scientific texts", which is much greater, and the shares of different genres represented in the libraries at Sippar and Kalhy, he adds the following comment (my translations):
[These numbers] mean that the mythic-epic texts and the hymns and prayers, which are categorised as literary texts, constitute 9 to 18% of the total corpus. When adding the pseudo-historical texts, which we already counted as literary texts, the share of literary genres may increase to 10 to 20%.
Note that many cuneiform tablets and tablet fragments in museum collections have not yet been published or even categorised. (The British Museum alone owns 130.000 clay tablets and tablet fragments.)
The available text corpus in the Akkadian language is the second biggest of the available corpora in ancient languages; only the Greek text corpus is bigger. However, the corpora in Akkadian and Sumerian (and other languages that used cuneiform) is much more likely to grow significantly than those in Greek or Latin.
When combining Jursa's information with that provided by Streck, the conclusion seems to be that literary texts may constitute at most 20% of the available canonical texts, which constitute less than 20% of the total corpus (i.e. roughly 20% minus the share of "monumental texts").
In other words, literary texts constitute less than 4% of the total corpus.