In Matthew 26:28, did Jesus say, "This is my blood of the new covenant"? Or did he say, "This is my blood of the covenant'? The contest, in Greek, is between τὸ τῆς καινῆς and τῆς. The external evidence - as presented in the apparatus of Wayne Mitchell's The Greek New Testament, 4th edition - shows that representatives of multiple text-types support τὸ τῆς καινῆς or τῆς καινῆς: the Byzantine text finds allies in A, C, D (without the τὸ), E, F, G. H. K, M, S, U, W, Γ, Δ, Π Ω 074vid f1 f13 28 205 565 579 597 700 892 1006 1071 1241 1243 1342 1505 1582 Lect the Old Latin and Vulgate, the Peshitta, Palestinian Aramaic, Sahidic and Bohairic versions (except for one Bohairic copy, and Schenke's Middle Egyptian), Armenian, Ethiopic, and part of the Old Georgian version. The Byzantine reading also has support from Irenaeus (in Latin), Origen (in Latin), Theophilus of Alexandria, Theodoret, Jerome, and Augustine.
P45 (damaged, but with space-considerations taken into account) and P37 agree with Vaticanus and Sinaiticus (and 019 035 038 33) on the shorter reading. Irenaeus (as preserved in Armenian) agrees with the shorter reading, and so do Cyprian and Cyril.
Both readings are clearly ancient.
Looking at the parallel in Mark 14:24, the longer reading is paralleled word for word in the Byzantine Text. Meanwhile, the passage without "new" is supported by Vaticanus, Sinaiticus, and L D P W Z Θ Ψ and Old Latin Codex Bobbiensis.
Metzger proposed that the longer reading in Mt. 26:28 originated via a harmonization to Luke 22:20. I propose, however, that something else has affected the text of Matthew 26:28. And it wasn't Marcionism. It could be imagined that Marcion or a Marcionite created the shorter reading because to Marcion, Jesus Christ did not introduce a new covenant; to Marcion, the one true God had nothing to do with the covenant of the Law.
Metzger asserted that if καινῆς had been present in the original text of Matthew 26:28, "there is no good reason why anyone would have deleted it." Some might insist that a Marcionite's theology would be, to him, a reason to delete it. But can a Marcionite's influence upon the Alexandrian text of Matthew have been so strong? Marcion himself only accepted his own edited text of the Gospel of Luke. So the idea that Marcionism was a factor seems unlikely.
But the flimsiness of an arrow thrown at the shorter reading does not really prove the strength of the shorter reading. If the shorter reading is regarded as original, then the text of Matthew 26:28 must have been harmonized to Luke 22:20 in multiple transmission-streams (affecting the Byzantine Text, the Old Latin and Vulgate, the Sahidic, the Sinaitic Syriac, Armenian, Ethiopic, and Slavic versions). Neither Lachman nor Tregelles seems to have thought that was a plausible option.
A less sinister mechanism than Marcionism seems to have been at work in the Alexandrian text of Matthew 26:28: simple parablepsis. A scribe beginning with τῆς καινῆς before διαθήκης could skip καινῆς by accidentally jumping from the -ῆς in τῆς to the -ῆς at the end of καινῆς. Perhaps slightly facilitating the omission of καινῆς was the influence of scribes' recollection of Exodus 24:8 as written in the Septuagint, where Moses "took the blood, sprinkled it on the people, and said, 'This is the blood of the covenant which the LORD has made with you according to all these words.'" There is no καινῆς in Exodus 24:8, the passage that Christ's words in Matthew 26 reflect. Contrary to Metzger's assertion that "there is no good reason" for a deletion in Matthew 26:28, it is easy to see that a mechanism of deliberate harmonization (to Exodus 24:8) and a mechanism of accidental omission could both contribute to the creation of the shorter reading. (Whenever an accidental omission occurs, aren't observations about the lack of motive superfluous?)
A wild card should not be overlooked: the word τὸ before τῆς καινῆς in the Byzantine Text. Non-Greek scribes might not have bothered with this; Greek scribes may have naturally added τὸ, regarding the resultant reading to be a slight stylistic improvement not affecting the meaning of the text. (Conversely, Alexandrian scribes might have considered it unnecessary.) This detail need not be resolved to maintain the conclusion that καινῆς was part of the original text of Matthew 26:28.