M6 and M5 Ultra Geekbench 7 Results--Apple's Claims Hold Up, the $6,799 36-Core Mac Studio Leads the M3 Ultra by 35 Percent, and the M6 Pro Score Is Likely Fake

The first Geekbench 7 results for Apple's M6 and M5 Ultra surfaced on September 15 and September 18, days before September 22, the date Apple's Mac Studio announcement gave for the Mac Studio with M5 Ultra to begin arriving, with the Mac mini with M6 on the same schedule, and they say the performance claims Apple made on August 25 were, if anything, conservative. A third result, for an M6 Pro that Apple has never announced, is probably a fake. Each CPU score below is one unverified upload, and the graphics figures are the best MacRumors had seen by September 18, so read them as early runs, not a verdict.

The M6 scores what Apple said it would

MacRumors reported the M6 result from the Geekbench 7 database on September 15: 4,071 single-core and 22,783 multi-core, against average M5 scores of 3,641 and 17,948, which it works out to about 12 percent and 27 percent faster. Apple's August 25 announcement claimed "the world's fastest single-threaded performance" and "up to 1.2x faster multithreaded performance" than the M5, and MacRumors says the 4,071 tops every Mac and PC in the database, including the M5 Max at 3,731 and Qualcomm's Snapdragon X2 Elite Extreme at 3,396. AppleInsider read the same listing, a "Mac18,5" with a 4.78GHz base frequency, against the Mac the new mini replaces: 24 percent faster single-core and 48 percent faster multi-core than the M4 Mac mini. Apple's own footnote says its comparison used a preproduction mini with 32GB against a 14-inch MacBook Pro with M5, and MacRumors notes that its M5 averages come from that MacBook Pro as well.

The 36-core M5 Ultra leads the M3 Ultra by a third

The M5 Ultra result, also from September 15, is from the top chip: a 36-core CPU at 3,774 single-core and 52,516 multi-core. MacRumors puts the 32-core M3 Ultra's averages at 2,920 and 39,059, so the gain is about 30 percent single-core and 35 percent multi-core, against Apple's claim of up to 1.25x and 1.3x. Apple never released an M4 Ultra, so the M3 Ultra is the chip this one replaces. AppleInsider notes that the tested Studio had 256GB of memory, a $10,799 configuration, and that the multi-core score is about six times the original M1's 8,434. The 36-core CPU is a step up from the base chip: Apple prices the Mac Studio with M5 Ultra from $5,499, and MacRumors says the 36-core model starts at $6,799.

The graphics scores came three days later

On September 18 MacRumors reported the first GPU results. The M5 Ultra posted a Metal score of 366,744, 59 percent above the M3 Ultra's 230,420 average, and an OpenCL score MacRumors calls "roughly in line with NVIDIA's GeForce RTX 4080"; Apple had claimed up to 40 percent faster graphics than the M3 Ultra. The M6 posted 93,217 in Metal, 34 percent above the M5's 69,563 average, which puts it below the M4 Pro at 107,354 and above the M2 Pro at 77,176 in MacRumors' table. Apple framed its M6 graphics claim differently, as "a nearly 30 percent increase in peak GPU compute for AI" over the M5, so the two figures measure different things and do not compare.

The M6 Pro result is probably fake

A third listing, dated September 18, turned up on Reddit: a "Mac18,6" with an 18-core "Apple M6 Pro" scoring 4,150 single-core and 37,565 multi-core, per AppleInsider, which noted that the multi-core figure sits a little above the 18-core M5 Pro's. On September 19 MacRumors reported that Geekbench's creator, John Poole, had found "internal inconsistencies" in the result and believes it might not be legitimate. That squares with what Bloomberg's Mark Gurman reported in July, in MacRumors' summary: no M6 Pro, M6 Max or M6 Ultra, a base M7 in the first half of 2027, and M7 Pro and M7 Max chips in the second half.

We have been here before

Independent numbers arriving before the machine is the argument we had in September 2003, when PC Magazine matched a dual 2GHz Power Mac G5 against a dual 3.06GHz Xeon Dell 650 it called "similarly priced" and we configured the Dell at more than $1,500 more; our follow-up of September 29 had Ziff-Davis revising the article but not the phrase. Then the magazine's price was the thing that was wrong; now Apple's claims are the ones holding up, and the suspect is a stranger's upload. A few runs from machines that had not yet reached customers are a good sign and not a review; we will post our own numbers when a mini and a Studio reach the bench, and until then Apple's 1.2x and 1.3x look like the safe end of the range.