M5 Ultra… Apple Wasn’t Messing Around
Credibility score: 69/100 — Mostly Credible. Mixed credibility - some claims are solid, others need verification.
BSmeter analyzed "M5 Ultra… Apple Wasn’t Messing Around" and rated it 69/100 for credibility (a BS score of 31/100 — mostly credible), on 2026-09-22. Its weakest claim — "Claims a 'performance bump' and 'price bump' using 'apparently' twice." — scored 45/100 and was flagged as volume game. 28 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 28 claims analyzed: 0 scored under 40, 6 between 40 and 69, and 22 at 70 or above.
Claims analyzed
Apple's M5 Ultra performance claims — presented as a question of 'real or marketing' — No Frame (75/100)
At 0:00
He's just laying out Apple's claims and then immediately questioning them. That's the setup. 🔥
Why this score: The speaker introduces Apple's stated performance improvements for the M5 Ultra (4.3x faster AI, 2x faster storage, 1.3x faster CPU) and then directly asks if these are 'real or just marketing.' This is a straightforward setup for his own testing, not a claim itself. He's being upfront about the source of the numbers and his intent to verify.
Original quote: “Well, the M5 Ultra is finally here, and it's right here on my desk, right next to the M3 Ultra. Apple says it has local AI performance up to 4.3 times faster than the previous generation. Two times faster storage, 1.3 times faster CPU speeds. Sounds pretty good. But is that real or is that just…”
M5 Ultra memory bandwidth and starting price — straight from the spec sheet — No Frame (75/100)
At 0:18
He's citing the spec sheet for the memory bandwidth and the starting price. No tricks here, just the numbers. 😈
Why this score: The speaker states the M5 Ultra's memory bandwidth (1.2 TB/s) and its starting price ($5499) directly from the spec sheet. He also explicitly says he's going to test the bandwidth. This is factual information presented without any manipulative framing, setting the stage for his review.
Original quote: “Well, according to the spec sheet, the new memory bandwidth is 1.2 terabytes per second. I'm going to test that out, of course. M5 Ultra is what I have here. That one starts at $54.99. And if you”
The speaker states the price of the configured M5 Ultra is $14,299. — No Frame (75/100)
At 0:46
He's just laying out the cost of his specific setup. No trickery here, just a number. 😈
Why this score: The speaker is transparently stating the price of the M5 Ultra configuration he's discussing, which is a straightforward factual statement. No hidden agenda, just the brutal cost.
Original quote: “So, I thought I'd give you the price, which comes to 14,299.”
Claims a 'performance bump' and 'price bump' using 'apparently' twice. — Volume Game (45/100)
At 0:51
He says 'apparently' twice in one breath, trying to sound casual while still making the claim. That's not caution, that's a quiet dodge. 😈
Why this score: The speaker uses 'apparently' twice to introduce both the performance and price bumps. While it softens the claim, it also allows him to state it without full commitment, a classic volume game where the claim is made loudly, but the responsibility is quietly retracted.
Original quote: “Not only do you get a performance bump apparently, but you also apparently get a price bump quite a bit.”
Compares M3 Ultra (47) to M5 Ultra (60.2) Speedometer scores, then pivots to M6 being faster. — Volume Game (45/100)
At 1:52
He gives you the M5 score, then immediately undercuts it by saying the M6 is 'much faster.' That's not a comparison, that's a setup and a punchline. 😈
Why this score: The speaker presents a direct comparison of Speedometer scores between the M3 Ultra and M5 Ultra, showing a clear improvement. However, he immediately pivots to mention the M6 chip, stating it will be 'much faster,' effectively diminishing the M5's achievement right after highlighting it. This is a classic volume game: give with one hand, take back with the other.
Original quote: “something, it's overkill. Here's a speedometer score. 47 is the M3 Ultra score. We got 60.2 on the M5 Ultra. I just did a M6 mini review. And yeah, it destroys that score by quite a lot. Now, this is the M5 chip. That's the M6 chip. So, a single core of that is going to be much faster in the M6…”
States M3 Ultra took 71.7 seconds for a compilation, M5 Ultra took 52.4. — No Frame (75/100)
At 2:22
He's just giving the raw compilation times for each chip. Straightforward data, no spin. 😈
Why this score: The speaker presents specific, measurable compilation times for both the M3 Ultra and M5 Ultra. This is a direct, factual comparison of performance data, without any apparent rhetorical framing or manipulation.
Original quote: “compilation. We've got 71.7 seconds on the M3 Ultra, 52.4”
Claiming M5 Ultra is the fastest he's ever seen for this task. — Personal Story (60/100)
At 2:30
His personal 'fastest ever' doesn't make it a universal truth, but it's his experience. 😈
Why this score: The speaker is expressing a personal observation about the M5 Ultra's speed, stating it's the fastest he has personally witnessed. This is an anecdotal claim based on his own experience, not a universally verifiable benchmark.
Original quote: “on I think that's the fastest I've ever seen it on the M5 Ultra.”
Stating M5 Ultra storage is 'more than two times faster' than M3 Ultra, citing specific benchmarks. — No Frame (75/100)
At 3:03
He's got the numbers right there: 6.5GB/s vs. 14.8GB/s read. The math checks out. 😈
Why this score: The speaker provides specific benchmark numbers for storage speeds (M3 Ultra: 6.5 GB/s read, 2.7 GB/s write; M5 Ultra: 14.888 GB/s read, ~20 GB/s write). The claim that it's 'more than two times faster' is directly supported by these figures, which are displayed on screen, making it a verifiable statement.
Original quote: “I've never seen it that fast. Shh. It's okay. It'll get the job done. Storage is supposed to be twice as fast. And I just ran Amorphous Disc Mark M3 ultra sequential speed. So, this is like copying a large file back and forth. 6,500 megabytes per second. So, 6.5 GB per second for read and 2.7 for…”
Asserting 'two times faster' is a 'conservative label' and that the M5 Ultra is 'actually faster than that'. — Confidence Mismatch (45/100)
At 3:46
He's saying Apple's marketing is 'conservative' based on his own tests. Bold claim for a mortal. 😈
Why this score: The speaker claims Apple's 'two times faster' marketing is conservative, implying the M5 Ultra is significantly more performant than even Apple suggests. While his benchmarks show impressive results, declaring a marketing team's label 'conservative' based on limited, specific tests is a subjective interpretation rather than a universally proven fact. He's confident in his assessment, but it's still an opinion on marketing strategy.
Original quote: “So yeah, two times faster is a conservative label by marketing team. It's actually faster than that. This matters for AI as well. When you're loading a 140 GB model, it's going to read it off the disc much faster than this one. Editor Alex here. This number turned out to be way too low and I…”
M5 Ultra's GPU cores have dedicated neural accelerators, unlike M3 Ultra. — No Frame (75/100)
At 4:55
He's laying out the technical specs for the M5 Ultra — sounds like a genuine architectural upgrade. 😈
Why this score: The speaker is detailing a specific hardware improvement in the M5 Ultra chip compared to the M3 Ultra, stating that each GPU core now includes a dedicated neural accelerator for AI tasks. This is a technical specification that would be part of the chip's design and public announcements, not a rhetorical trick. It's a straightforward description of a product feature.
Original quote: “But it's not just an incremental improvement. It's not a small improvement because in the M3 Ultra, in the M3 generation, the GPU cores were just GPU cores. They were just regular old cores. But now with the M5 generation, we have for every GPU core, there's a neural accelerator that lives inside…”
Llama CPP shows 'has tensor = true' on M5 Ultra, 'false' on M3 Ultra. — No Frame (75/100)
At 5:18
He's showing a direct, verifiable software output difference between the chips. That's called evidence, mortals. 😈
Why this score: The speaker provides a direct, observable software output ('has tensor = true/false') as evidence for the M5 Ultra's new capabilities. This is a specific, testable claim about how software interacts with the hardware, not a rhetorical device. It's a demonstration, not a manipulation.
Original quote: “Now, check this out. On the M3 Ultra, Llama CPP starts up and says has tensor equals false. Same exact build on the M5 Ultra, has tensor is true. So, Llama CPP supports metal. It's actually one of the first things Llama CPP supported. And now it has specific path that the software takes for the M5…”
CPU benchmark shows lower memory bandwidth than Apple's claim, then explains why. — No Frame (75/100)
At 7:09
He's showing the numbers, then immediately explaining the discrepancy. That's called being upfront, mortals. 😈
Why this score: The speaker presents benchmark results that are lower than Apple's claimed specifications for the M3 Ultra, but then immediately provides a clear and logical explanation for the difference (the test is CPU-related, not GPU). This is transparent and not misleading.
Original quote: “And boom. Now the M3 Ultra 819 GB per second is what Apple claims. We're getting 312.9 gigabytes per second for the Triad and 583.6 gigabytes per second on the M5 Ultra. And the reason for the lower number is because stream is actually a CPU related test. So, it's running on the CPU memory…”
GPU benchmark results for M3 and M5 Ultra, aligning with Apple's claims. — No Frame (75/100)
At 7:44
He's showing the numbers, then immediately explaining the discrepancy. That's called being upfront, mortals. 😈
Why this score: The speaker presents benchmark results for GPU memory bandwidth that are close to Apple's claimed numbers (89-90% for M3 Ultra, 85-86% for M5 Ultra). This is a straightforward presentation of data with a reasonable comparison.
Original quote: “Apple says 819. So that's about 89 90%. And the M5 Ultra 1,039 GB per second out of 1.2 TB that's about 8586%.”
Asks if token generation will be 1.4x faster, setting up the next test. — No Frame (75/100)
At 8:01
A direct question, setting up the next test. No tricks here, just curiosity. 😈
Why this score: The speaker poses a direct question that leads into the next segment of his testing, which will directly address this question. It's a clear transition and not a manipulative framing technique.
Original quote: “Does that mean that token generation is going to be 1.4 times faster?”
Deepseek V4 Flash token generation: M3 Ultra at 37 tokens/sec, M5 Ultra at 53 tokens/sec, 1.5x faster. — No Frame (75/100)
At 8:10
He's showing the actual performance numbers and the speedup. The data speaks for itself. 😈
Why this score: The speaker provides specific, measured results for token generation on a large language model (Deepseek V4 Flash) for both the M3 Ultra and M5 Ultra, and calculates the speed increase. This is a direct presentation of benchmark data.
Original quote: “All right, let's start big. Deepseek V4 Flash is 284 billion parameters. By the way, this is a pretty big model and it's pretty popular now. And we're getting 37 tokens per second on the M3 Ultra, 53 on the M5 Ultra. And that is about one and a half times faster. Kind of lines up, right?”
Compares M3 Ultra to M5 Ultra performance, notes Apple's higher claim. — No Frame (75/100)
At 8:52
He's giving numbers and then immediately flagging the discrepancy with Apple's own claims. Straightforward reporting. 😈
Why this score: The speaker presents specific performance numbers for the M3 and M5 Ultra, then directly contrasts his findings with Apple's official claims, acknowledging the difference. This is transparent and allows the viewer to see the data and the potential variance.
Original quote: “483 tokens per second on the M3 Ultra and 1485 on the M5 Ultra. That's three times. Now, we're talking, but Apple said four.”
Updates M3 Ultra performance and compares it to M5 Ultra on a specific model. — No Frame (75/100)
At 9:03
More numbers, more specific models, and he's even re-tested it himself. This is what mortals call 'evidence.' 😈
Why this score: The speaker provides updated performance figures for the M3 Ultra and compares them to the M5 Ultra using a specific 120 billion parameter model (GPTOSS120B). He notes the testing environment (LM Studio vs. Pure Lama CPP) and states he re-tested it, adding transparency to his data.
Original quote: “Now, different models behave differently. And you might remember this one from the M5 Max video that I did a few months ago. And at that time, the M3 Ultra was king at 82 tokens per second. That's GPTOSS120B, 120 billion parameter model. Now, that was in LM Studio, not Pure Lama CPP. Now, we're…”
Demonstrates M3 Ultra vs. M5 Ultra performance with large context window. — No Frame (75/100)
At 9:46
He's showing a real-world test with massive context. The numbers are right there, no smoke and mirrors. 😈
Why this score: The speaker describes a specific test scenario involving a large context window (64,000 tokens already in context, adding 32,000 more) and provides the exact processing times for both the M3 Ultra and M5 Ultra, then calculates the speed difference. This is a clear, demonstrable comparison.
Original quote: “Now, watch this. With 64,000 tokens already in context, I gave it 32,000 more. That took 80 seconds on the M3 Ultra and 56 seconds on the M5 Ultra. So, that's 1.4 times.”
Comparing M3 Ultra to M5 Ultra token generation — straightforward numbers. — No Frame (75/100)
At 10:30
He's just laying out the numbers, M3 vs M5. No tricks, just data. 😈
Why this score: The speaker is presenting a direct comparison of token generation speeds between two specific chip models, M3 Ultra and M5 Ultra, with clear numerical values. It's a factual comparison without any apparent rhetorical manipulation.
Original quote: “No, I took science class. cuz I know what density is. Token generation on the M3 Ultra, 37 tokens per second, 54 tokens per second on the M5 Ultra. So that's 1.47 times. Basically, we're getting the same kind of uh scale on all kinds of models here.”
Claiming Apple's 'four times faster' prompt processing is accurate, based on his own test. — No Frame (75/100)
At 10:50
He's verifying Apple's claim with his own test. The numbers match up. Fine. 😈
Why this score: The speaker is directly testing Apple's claim of 'up to four times faster prompt processing' and providing his own measured results (4.2 times faster) that align with the company's statement. This is a direct verification, not a framing trick.
Original quote: “Now, prompt processing 430 on the M3 Ultra and on the M5 Ultra 1,800. Holy cow, that's kind of like four times, right? It's actually more than four times. I also gave them both a 14,000 token prompt, equivalent to about a handful of source files if you're doing code, just to see if it falls apart…”
Qualifying the 'four times faster' claim by showing it depends on prompt length. — Missing Context (45/100)
At 11:32
He's admitting the 'four times faster' is conditional, but only after selling the big number. That's a classic volume game. 🚩
Why this score: The speaker initially presents the 'four times faster' figure as a general improvement, then later clarifies that this performance gain is highly dependent on the length of the prompt. This is a classic 'Volume Game' where a bold claim is made, and then the necessary context or limitations are added quietly afterward, potentially diminishing the initial impact for those who don't listen closely to the caveats.
Original quote: “Now, don't get me wrong, I'm impressed. But before you get your credit card out, and it is going to be a big credit card bill, that four times needs a long prompt. At about 1,700 tokens, it's 3.4 times. It takes about 4,500 tokens to hit the four time scale, which shouldn't be a problem if you're…”
Downplaying efficiency gains by stating tokens per watt only increased by 12%. — No Frame (75/100)
At 12:04
He's just giving the efficiency numbers. No spin, just the raw data. 😈
Why this score: The speaker is providing specific power consumption figures for both chips during generation and idle, and then calculating the 'tokens per watt' efficiency gain. This is a direct presentation of data to support his conclusion about efficiency, without apparent rhetorical manipulation.
Original quote: “It's also not much more efficient. During generation, the chip reported about 45 watts versus 36 on the M3 Ultra. And that's GPU plus CPU, not wall power. So, tokens per watt only went up to about 12%. This is basically the machine idling 12 11 or 12 watts for the M3 Ultra and about eight or nine…”
Comparing M3 Ultra's 200 watts to M5 Ultra's 400+ watts under heavy GPU load. — No Frame (75/100)
At 12:30
He's showing the GPU history chart right there, the numbers are on display. No trickery, just raw data. 🔥
Why this score: The speaker is directly referencing the on-screen GPU history chart, which visually supports the power consumption figures he's quoting for both the M3 Ultra and M5 Ultra under 100% usage. It's a direct observation of the presented data.
Original quote: “or nine up to 10 watts on the M5 Ultra. However, during heavy intense GPU work, I got it pegging right now. Uh that's a GPU history chart right now on both of them all the way to 100% usage. And the M3 Ultra is hitting almost 200 watts. The M5 over 400 watts. That is crazy.”
Claiming the M5 Ultra is 'definitely much much more audible' and can be heard through his microphone. — Personal Story (60/100)
At 12:53
He's telling you what he hears and what his mic picks up. It's his experience, not a scientific claim. 👂
Why this score: The speaker is describing his personal auditory experience with the M5 Ultra, stating it's 'much much more audible' and suggesting the audience might hear it through his microphone. This is a subjective observation based on his immediate environment and equipment, not a universally verifiable fact.
Original quote: “uh the M5 Ultra first. Is definitely much much more audible. You can probably hear it from my microphone. That's all M5 Ultra right now.”
Stating the M5 Ultra's 'bright orange' color on the display indicates it's heating up 'quite a bit more'. — No Frame (75/100)
At 13:08
The visual evidence is right there on the screen, showing the M5 Ultra's temperature reading in a more intense color. It's a direct observation. 🌡️
Why this score: The speaker is making a direct visual observation from the on-screen display, where the color coding (bright orange) is typically used to indicate higher temperatures. This is a straightforward interpretation of the visual data presented.
Original quote: “I'm hearing the M5 Ultra. And look at the temperatures. You can just tell from the bright orange of the M5 Ultra that it's heating up quite a bit more.”
Stating token generation is 1.5x faster and prompt processing 2-4x faster, which coding agents will 'feel'. — No Frame (75/100)
At 14:08
He's giving specific performance metrics and a direct benefit. It's a clear statement of observed improvement. 🚀
Why this score: The speaker is providing specific performance metrics (1.5x faster token generation, 2-4x faster prompt processing) which are presented as direct results of his testing. While he doesn't show the full test, these are quantitative claims that would typically be backed by benchmarks, and he implies further testing is coming.
Original quote: “bottom line, token generation one and a half times faster. Nice upgrade. Prompt processing two to four times faster. For coding agents, you will feel that.”
Previewing a test where the M5 Ultra outputs 134 tokens/second against the M3 Ultra's 75 tokens/second. — No Frame (75/100)
At 14:19
He's giving a direct comparison from a specific test, showing the numbers side-by-side. It's a clear benchmark. 📊
Why this score: The speaker is citing a specific benchmark result from a test he conducted ('eight requests at once'), providing concrete numbers for both the M5 Ultra and M3 Ultra. This is a direct, measurable comparison presented as evidence for the performance claims.
Original quote: “just to give you a little preview with eight requests at once. This thing puts out 134 tokens per second against M3 Ultra 75.”
Just a standard outro — no claims, just a call to action and a preview. 😈 — No Frame (75/100)
At 14:30
He's just wrapping up the video and promoting future content. Nothing to dissect here. 🔥
Why this score: This is a straightforward outro, typical for YouTube videos. The speaker is simply encouraging viewers to subscribe and watch other videos, without making any factual claims or using manipulative language.
Original quote: “going to dive a little deeper into MLX versus Llama CPP and 8bit models. So stay tuned for that. Make sure you subscribe so you don't miss it. And if you want to check out the M5 Max and its capabilities, that video is right over here. Thanks for watching and I'll see you in the next one.”
See the full analysis with timestamps →