I was SO WRONG... MBP M5 MAX vs M4 MAX for Local AI TESTED 🤔
Credibility score: 50/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "I was SO WRONG... MBP M5 MAX vs M4 MAX for Local AI TESTED 🤔" and rated it 50/100 for credibility (a BS score of 50/100 — mixed credibility), on 2026-04-18. Its weakest claim — "YouTubers said M5 Max 20% faster, he predicted 2.5x" — scored 50/100 and was flagged as opinion. 6 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 6 claims analyzed: 0 scored under 40, 3 between 40 and 69, and 3 at 70 or above.
Claims analyzed
YouTubers said M5 Max 20% faster, he predicted 2.5x — Opinion (50/100)
At 0:12
Dropping '2.5x faster' prediction like it's gospel when it's pure speculation — love the confidence, hate the lack of receipts 💀🤔
Why this score: Pure opinion and prediction, not a factual claim.
- Speaker admits it's his own prediction vs. others' 20% estimate
- No evidence provided for any figure; sets up untested hype
- M5 Pro reference is teaser, actual tests later (and community notes benchmarks issues anyway)
Original quote: “A lot of YouTubers out there, I [0:17] said it's 20% faster. Me, I predicted it [0:20] could be two and a half times faster, [0:21] but we already seen the M5 Pro. Is it [0:23] two times faster than the M4 Max?”
Went to Apple HQ, got M5 Max, did benchmarking — Personal Story (65/100)
At 2:30
Went to Apple HQ like it's a corner store for chips? Bold flex but M5 Max isn't even out yet in stores 💀👀 — sounds like store demo spin, not HQ heist.
Why this score: Personal anecdote about testing process — can't verify exact events but M5 Max availability is limited as of April 2026 to late 2025 launches and mid-2026 rumors, no public HQ access for random benchmarking. *Plausible if Apple Store*, but HQ claim amps drama without proof. Score reflects unverifiable testimony with contextual red flags.
Original quote: “we didn't stop there. We went ahead. [1:46] We went to Apple HQ to find out for [1:49] certain. Went ahead and got the M5 Max [1:51] and we did some benchmarking.”
Qwen 3.5 27B 4-bit: 22-23 tokens/sec on M5 Max — Solid (75/100)
At 3:14
22-23 t/s on Qwen 3.5 27B 4bit out-the-box? I'm mad that's believable for M5 — store benchmark delivering actual heat 😤✅🔥.
Why this score: Performance claim aligns with M5 Max's AI boosts; quantized 27B models like Qwen2.5 (likely 'coin' = Qwen) hit 20+ t/s on high-end Apple Silicon per benchmarks. Store tests feasible with dev tools. *Would love full logs*, but numbers track expected gains over M4.
Original quote: “We're doing coin 3.527B. [2:50] If you want to play along at home, 20 [2:52] 3.5 27B the 4bit edition just straight [2:55] out of the box. Boom. We're doing 22 to [2:58] 20 almost 23 tokens a second.”
M5 Max inference test: Wikipedia article on Appel with neural acceleration — OK (60/100)
At 3:46
Wikipedia on 'Appel' for AI benchmark? Either genius troll or auto-complete fail — neural accel sounds right but setup screams sketchy store vibes 🤔💻.
Why this score: Benchmark setup plausible for local AI inference, as M5 excels in Neural Engine tasks per specs. But 'Appel' likely typo/mispronunciation for Apple, and store tests are real but not controlled lab conditions. *Limited verification without raw data or video of screen* — mid score for demo legitimacy.
Original quote: “let's just jump in [2:11] ahead and look at a test over here. So [2:13] I'm going to jump in and we got [2:16] inference right here with the software [2:17] update using neural acceleration. We got [2:20] a Wikipedia article about Appel.”
M5 Max generated 100 tokens in 15.65 seconds — Solid (78/100)
At 3:46
15.65s for 100 tokens = ~6.4 t/s total? Wait, earlier 23 gen — math's close enough for live demo slop. Actually solid 👏📊.
Why this score: Timestamp verifiable if screen shown; equates to reasonable inference speed including prompt processing (earlier hyped gen rate). M5 > M4 by design for local AI. *Minor discrepancy in reported speeds typical of variable tests* — holds up.
Original quote: “100 tokens [3:00] were made only 100. We're just keeping [3:02] [4:20] it simple. Check out the processing and [3:03] boom. It says right there 15.65 seconds. [3:06] So you want to see how the M4 Max does”
Disabled parental controls for restricted words in Wikipedia article — Personal Story (70/100)
At 4:16
'Printal control' and disabling for Wikipedia swears? This demo's got more drama than the results 😂🛡️ — relatable tech sacrifice tho.
Why this score: Anecdote checks out as parental controls (likely 'Parental Control') often flag profanity, even in neutral articles like Apple Wikipedia page. Common in AI apps testing uncensored models. *No contradiction*, just quirky detail boosting authenticity.
Original quote: “Now I had to sacrifice myself [2:28] and disable restrictions. So I have to [2:31] disable restricted mode because the [2:34] Wikipedia article had some restricted [2:36] kind of words. So it's called printal [2:37] control this application.”
See the full analysis with timestamps →