Qwen3.8-27B & How to Serve it Fast
Credibility score: 56/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "Qwen3.8-27B & How to Serve it Fast" and rated it 56/100 for credibility (a BS score of 44/100 — mixed credibility), on 2026-09-17. Its weakest claim — "Qwen 3.8 Max is 'undeniably amazing' — a bold, subjective claim." — scored 45/100 and was flagged as loaded language. 40 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 40 claims analyzed: 0 scored under 40, 26 between 40 and 69, and 14 at 70 or above.
Claims analyzed
Qwen 3.8 Max is 'undeniably amazing' — a bold, subjective claim. — Loaded Language (45/100)
At 0:00
Calling a model 'undeniably amazing' is pure hype. 'Amazing' is an opinion, not a spec sheet. 😈
Why this score: The speaker uses 'undeniable' and 'amazing' to present a subjective assessment as objective fact. It's a strong emotional appeal to establish the model's perceived value without offering specific, quantifiable metrics at this point.
Original quote: “Okay, so last week, Qwen dropped the weights for Qwen 3.8 Max, and
it's undeniable that this model is pretty amazing at what it can do.”
The 2.4 trillion parameters make it unrunnable locally for 'very few people'. — Confidence Mismatch (45/100)
At 0:10
2.4 trillion parameters is a beast, but 'very few people' is a guess. How few? Got a number, or just vibes? 💀
Why this score: The speaker states with certainty that 'very few people' can run this locally. While the parameter count is massive and implies high resource demands, the exact number of people capable of running it is speculative without specific data on hardware ownership among potential users. It's a confident assertion without a clear basis.
Original quote: “But the key problem with this is its 2.4 trillion parameters.
So there are very few people that are gonna be able to run this locally.”
Claims 'everyone' was waiting for the Qwen 3.8-27B model — a broad generalization. — Loaded Language (45/100)
At 0:19
Oh, 'everyone' was waiting, were they? I've seen 'everyone' wait for the apocalypse, not a model. 🙄
Why this score: The speaker uses 'everyone' to imply universal anticipation for the Qwen 3.8-27B model. This is a generalization that exaggerates the sentiment, likely to build excitement and importance around the topic. It's a common rhetorical device to create a sense of shared interest.
Original quote: “What everyone was actually waiting for was the model that dropped on
Friday, and this is the Qwen 3.8-27B.
So in this video, I'm gonna look at the model and the benchmarks, et cetera,”
Calling Qwen 3.6-27B the 'darling model' — a subjective, emotional appeal. — Loaded Language (45/100)
At 0:37
He's using 'darling model' to inject emotional appeal — makes it sound like a universal truth, not just his opinion. 😈
Why this score: The term 'darling model' is subjective and emotionally charged, implying widespread affection and success without providing concrete, objective metrics or sources to back up this claim of popularity. It's a rhetorical flourish, not a factual statement.
Original quote: “Now, I've talked a lot on the channel about the Qwen 3.6-27B, that really in many ways, that was sort of the darling model of a lot of people doing local coding, running local agents, et cetera. And we've even looked at things like ThinkingCap, which was a fine-tune of that 3.6 model to basically…”
Claiming Meta 'possibly' rushed out Glimmer 30B due to Qwen 3.8 — pure speculation. — Confidence Mismatch (45/100)
At 1:15
He's speculating on Meta's motives with zero evidence. 'Possibly' is doing a lot of heavy lifting for a claim about corporate strategy. 💀
Why this score: The speaker is attributing a motive ('trying to rush this out') to Meta's actions based purely on his own conjecture. There's no internal knowledge or public statement cited to support the idea that Meta's release schedule was dictated by Qwen's upcoming model. It's a guess presented as an informed observation.
Original quote: “And I made the comment at the time that they were possibly just trying to rush this out, knowing that the 3.8 27B was on the way and was likely to drop any day after that.”
Confirming his 'pretty true' speculation about Meta's rush, citing Qwen's own benchmarks. — Confidence Mismatch (45/100)
At 1:25
He's using Qwen's benchmarks to 'prove' his earlier speculation about Meta's motives. Benchmarks don't reveal corporate strategy, mortal. 🔥
Why this score: The speaker uses Qwen's benchmark results to retroactively validate his earlier speculation about Meta's rushing. While the benchmarks might show Qwen 3.8 performing better, this doesn't inherently prove Meta released Glimmer 30B in a hurry due to Qwen. Performance metrics are not evidence of a competitor's internal decision-making process.
Original quote: “Well, sure enough, that is pretty true. if we look at the benchmarks that actually came out, for this from Qwen, we can see that this model is doing substantially better than Meta's Muse Glimmer model, in here.”
Reiterating his 'pretty much true' claim about Glimmer being surpassed, again citing Qwen's own benchmarks. — Confidence Mismatch (45/100)
At 2:13
He's circling back to his own 'comments' and using Qwen's benchmarks to validate them. That's not proof, that's self-affirmation. 😈
Why this score: The speaker is once again using Qwen's self-published benchmarks to confirm his prior 'comments' about Muse Glimmer being surpassed. While the benchmarks might show Qwen's model performing better, relying solely on the developer's own benchmarks to confirm one's own prior statements, especially when those statements involved speculation about a competitor's motives, is a form of confirmation bias rather than objective proof.
Original quote: “So if we come in and look at Qwen's own benchmarks, You can see that the comments that I made back then about the Muse Glimmer were pretty much true, that this model pretty much surpasses Muse Glimmer.”
Qwen 3.8 is a "substantial bump" over 3.6 in "pretty much every area" — a confident, vague claim. 😈 — Confidence Mismatch (45/100)
At 2:42
He says 'substantial bump in pretty much every area' but shows no data, no specific metrics. That's not a claim, it's a feeling. 💀
Why this score: The speaker makes a very broad, positive claim about Qwen 3.8's performance improvement over 3.6 without providing any specific benchmarks, numbers, or areas of improvement. It's a confident assertion without the data to back it up, leaving the audience to take his word for it.
Original quote: “But back to Qwen 3.8.
So if we look at this, it's a substantial bump over 3.6 in
pretty much every area here.
So if you were happy using 3.6, you're definitely gonna get a
nice drop-in replacement here.”
Qwen 3.8 beats Opus 4.6 Max in computer use, then immediately qualifies it. 🚩 — Volume Game (45/100)
At 2:56
Claims Qwen beats Opus, then immediately says Opus wasn't 'heavily focused' on those areas. That's not a win, it's a rigged game. 😈
Why this score: The speaker makes a strong claim that Qwen 3.8 beats Opus 4.6 Max in 'computer use and stuff like that,' but then immediately undermines it by stating that previous Opus models weren't 'heavily focused' on those areas. This is a classic volume game: make a bold claim, then quietly retract or qualify it in the same breath, hoping the initial impact sticks.
Original quote: “On top of that, they seem to have put a lot of effort into the vision performance
in here, and we can see even things like computer use and stuff like that.
This is actually beating out, Opus 4.6 Max which to be fair, those previous
Opus models weren't heavily focused as much on computer use and…”
Qwen 3.8 scores 52 on Artificial Analysis, 'not far behind' GLM 5.2 (53) and 'way ahead' of Qwen 3.6. 😈 — No Frame (75/100)
At 3:15
He's citing specific scores from 'Artificial Analysis' and comparing them directly. Finally, some numbers. 🔥
Why this score: The speaker provides a specific source ('Artificial Analysis') and concrete scores (Qwen 3.8 at 52, GLM 5.2 at 53, Qwen 3.6 unspecified but 'way ahead'). This is a straightforward presentation of data, allowing the audience to verify or interpret the comparisons themselves. No obvious framing tricks here.
Original quote: “So just as I've been recording, Artificial Analysis has released their intelligence
index scores for this Qwen 3.8/27B model.
And I gotta say, the results are pretty insane.
This is scoring 52, Which puts it not far behind models like the GLM 5.2 at 53,
DeepSeek V4 Pro, And also just way ahead of…”
Qwen 3.8 beats GLM 5.2 and 'some' GPT 5.6 models on the agentic index. 💀 — Missing Context (45/100)
At 3:46
Beating 'some' GPT 5.6 models? Which ones? That's not a win, that's a vague flex. 😈
Why this score: The speaker claims Qwen 3.8 beats 'some of the GPT 5.6 models' on the agentic index. This is a classic case of missing context. Without specifying *which* GPT 5.6 models were beaten, the claim is vague and prevents any meaningful comparison or verification. It sounds impressive but lacks the crucial details that would make it a solid piece of evidence.
Original quote: “The other thing is when we look at the agentic index here,
this is actually beating out GLM 5.2, beating out some of the GPT 5.6 models.
Again, this is pretty insane for a model that you can run locally and
get a decent token speed out of it.”
Crediting Blackfrost AI as 'one of the first' — a specific, verifiable claim. — No Frame (75/100)
At 5:08
He's naming a specific team for a specific action. That's a direct claim, not a trick. 🔥
Why this score: The speaker attributes the action of creating 'abliterated versions' to the Blackfrost AI team, stating they were 'one of the first groups to do this.' This is a straightforward, verifiable statement of fact, not a rhetorical device.
Original quote: “so the Blackfrost AI team was one of the first groups to do this but”
Boasting 96GB VRAM and 'no problems' loading the model — a flex with missing context for most users. — Missing Context (45/100)
At 6:11
He's got 96GB of VRAM and says 'no problems.' That's not a flex, it's a taunt for anyone without a small data center. 💀
Why this score: The speaker highlights his 96GB of VRAM and states he has 'no problems being able to load the model.' While true for his setup, this statement lacks context for the average user, who would likely struggle with such a large model on consumer-grade hardware. It's a personal experience presented without acknowledging the significant barrier to entry for others, as community feedback also points out.
Original quote: “So I've got 96 gigabytes of VRAM here to play with, So in my particular case, I have no problems being able to load the model, right?”
Setting up the premise that quantization isn't everything. — No Frame (75/100)
At 6:41
He's just laying out his argument here — no tricks, just a setup for the demo. Fine. 😈
Why this score: The speaker is introducing the core idea that model performance isn't solely dependent on quantization, which is a reasonable premise for a technical discussion. No manipulative framing detected.
Original quote: “So first up, whatever model that you pick, you're gonna realize that it's not just”
Claiming the 'no thinking' model did a 'pretty good job' without showing the output. — Confidence Mismatch (45/100)
At 7:13
He says 'pretty good job' for the 'no thinking' model, but we haven't seen it yet. That's a judgment without evidence. 💀
Why this score: The speaker makes a subjective assessment ('pretty good job') of the model's output before actually displaying or detailing that output. This creates an initial impression without allowing the viewer to form their own judgment based on the evidence.
Original quote: “so it's done a pretty good job at, just basically thinking turned off completely.”
Highlighting a 'quite different' website from 'low thinking' without showing the 'no thinking' version for comparison. — Missing Context (45/100)
At 7:20
He's showing the 'low thinking' output and saying it's 'quite different' — but we still haven't seen the 'no thinking' version to compare it to. Where's the first one? 😈
Why this score: The speaker is discussing a comparison between different 'thinking' levels of the model's output. However, by not showing the 'no thinking' version alongside the 'low thinking' version, the viewer lacks the immediate visual context to fully appreciate the claimed difference. This makes the comparison less impactful and relies on the viewer's memory or trust.
Original quote: “If we look at the same thing for reasoning on with low thinking, you can see we get quite a different website, right?”
Claiming 512 thinking tokens is 'reasonably consistent' after 'a few different times'. — Confidence Mismatch (45/100)
At 7:42
He says 'reasonably consistent' after 'a few different times.' That's not data, mortal — that's a hunch with a spreadsheet. 💀
Why this score: The speaker uses the phrase 'reasonably consistent' based on 'a few different times' of running the test. This level of confidence is not fully supported by the vague quantification of 'a few different times,' which lacks statistical rigor or specific numbers of runs to establish true consistency. It's an assertion of reliability without robust evidence.
Original quote: “So here, if we look at the code, we can see we got 512 thinking tokens. And this seems to be reasonably consistent when I've run it a few different times.”
Claiming 'medium thinking' used fewer tokens than 'low thinking' after 'four or five times'. — Confidence Mismatch (45/100)
At 8:19
He's 'not sure why' it used less, but he's confident after 'four or five times.' That's not an explanation — that's a shrug with a small sample size. 😈
Why this score: The speaker presents a counter-intuitive result (medium thinking using fewer tokens than low thinking) and attributes it to having run the test 'four or five times.' While he acknowledges uncertainty ('I'm not sure why'), the small number of runs doesn't provide strong statistical evidence for this unexpected outcome, nor does it offer a robust basis for drawing firm conclusions about the model's behavior.
Original quote: “So you would presume here that this is gonna actually be using more thinking tokens, but actually it ends up using less. and I've run this sort of four or five times.”
Model 'goes nuts' and runs out of tokens on 'x-high' thinking. — A claim of observation. — No Frame (75/100)
At 8:43
He's just showing what happened with the model. No trick here, just a direct observation. 😈
Why this score: The speaker is describing a direct outcome of running the model with 'x-high thinking' and hitting token limits. This is a straightforward report of an observed event, not a manipulative framing.
Original quote: “And this is where the model goes nuts. You'll see here that I don't have a finished website to show you on this one, Because the issue here is that while I've limited it to 32K max tokens out, I've actually run out because I've spent 17,500 tokens on thinking.”
Running on 'x-high' thinking consistently leads to long waits and high token usage. — A claim of consistent observation. — No Frame (75/100)
At 9:01
He's sharing a consistent pattern he's observed across multiple runs. It's a direct report of his experience. 😈
Why this score: The speaker is detailing a consistent behavior he's observed with the model under 'x-high thinking' settings, specifically the high token consumption and processing time. This is presented as a direct, repeatable finding from his tests, not a rhetorical manipulation.
Original quote: “Now, I've run this multiple times trying to get one that actually finished, and I've had the thinking tokens be as high as, 22,000 thinking tokens. and if we look at them, it goes into insane detail in here. so while it actually produces a lot of the similar sort of colors and stuff as the, other…”
The 'x-high' thinking issue is consistent across FP8, 16-bit, and Unsloth models. — A claim of broad applicability. — No Frame (75/100)
At 9:45
He's extending his observation to different model versions. It's a direct report of his findings. 😈
Why this score: The speaker is stating that the observed behavior of 'insanely long amounts of thinking' on 'x-high thinking' is not limited to one specific model configuration but holds true across FP8, 16-bit, and Unsloth models. This is a direct report of his testing results, not a manipulative framing.
Original quote: “and this seems to be true whether I'm running on the FP8, which I've done these examples on, but I've also tried this on the 16-bit model and on the unsloth model. And, you'll see that for each of these, on xhigh thinking, you get insanely long amounts of thinking.”
Medium thinking reduces tokens from 11,000 to under 1,000 for the SVG test. — A direct comparison of token usage. — No Frame (75/100)
At 10:20
He's giving a direct numerical comparison. The numbers are right there. No trick. 😈
Why this score: The speaker is providing a direct, quantifiable comparison of token usage between 'x-high thinking' (11,000 tokens) and 'medium thinking' (under 1,000 tokens) for the SVG test. This is a factual report of his experimental data, not a framing technique.
Original quote: “But if we look at, the, medium thinking here, we're down from 11,000 down to just under 1,000 tokens.”
The 'ugly pelican' with no thinking tokens — a visual comparison. — No Frame (75/100)
At 10:39
He's just showing the visual result of less processing. It's a direct comparison, no tricks. 😈
Why this score: The speaker is presenting a visual outcome based on a technical setting. It's a straightforward demonstration of the model's performance without 'thinking tokens' versus with them. No hidden agenda here, just a clear example.
Original quote: “and we've got a sort of similar one, although we're definitely starting to miss some of the details there and sure enough, if we turn the reasoning off and just have a look at the straight one out with no thinking, we've got a pretty, ugly pelican.”
Declaring 'medium reasoning' as the sweet spot for model performance. — Personal Story (60/100)
At 10:56
He's calling 'medium' the sweet spot based on his own testing. It's a personal conclusion, not a universal law. 😈
Why this score: This is the speaker's personal assessment of the optimal setting based on his experience with the model. While it's a valid observation for his use case, it's presented as a personal finding rather than a universally proven fact. It's his opinion, not a claim about the model's inherent design.
Original quote: “So the sweet spot for me seems to be having the reasoning on medium.”
Claiming 'medium thinking' regressed on the dragon image, despite being good on the bicycle. — No Frame (75/100)
At 11:20
He's showing a clear visual regression on one element while another stays good. It's an honest observation. 😈
Why this score: The speaker is pointing out a specific visual outcome where increasing 'thinking' tokens actually made one part of the image worse while another remained stable. This is a direct, observable result from the demonstration, not a rhetorical trick. He's just calling it as he sees it on screen.
Original quote: “if we go up to medium thinking, you could even say that it's kind of regressed a bit on the dragon. Still pretty good on the bicycle, but not on the dragon there And”
Acknowledging 'xHigh' gets better results but at an 'insane' token cost. — No Frame (75/100)
At 12:01
He's stating a trade-off: better quality for a massive increase in tokens. It's a straightforward observation. 😈
Why this score: The speaker is presenting a clear cost-benefit analysis. He admits the 'xHigh' setting yields better results but highlights the disproportionate increase in token usage, which is a practical concern for deployment. This is a transparent assessment of the model's performance versus resource consumption.
Original quote: “so yes, the, the xHigh will get you better results, but the amount of tokens is just kind of insane.”
Speaker states bfloat16 version gets 30 tokens/second. — No Frame (75/100)
At 12:34
He's just stating his baseline test results. Straightforward enough. 🔥
Why this score: The speaker is providing a direct observation from his testing, establishing a clear benchmark for comparison. No obvious rhetorical tricks here.
Original quote: “So first off, I tested the basic bfloat16 original version of it.
and that was getting me, somewhere around sort of 30 tokens per second,”
FP8 version with speculative decoding hits 80-120 tokens/second with no quality loss. — No Frame (75/100)
At 13:13
He's giving his personal test results and subjective quality assessment. Fair enough. 😈
Why this score: The speaker is presenting his observed performance metrics and a subjective evaluation of quality. While 'didn't notice much of a quality difference' is subjective, it's presented as his personal experience, not a universal truth. The community comments also show varied results, but his own experience is his own.
Original quote: “to was the FP8 version of the model from Qwen themselves, And this with speculative
decoding was getting me around 80 to one hundred and twenty tokens per second.
And it works, very nicely.
I didn't really notice much of a quality difference compared
to the bfloat16 version.”
Abliterated models get stuck in loops, so he wouldn't bother with them. — Personal Story (60/100)
At 13:47
He's sharing his personal frustration with a specific model version. That's his experience, not a universal decree. 💀
Why this score: The speaker is recounting his personal experience with 'abliterated versions' of the model, noting a specific issue he encountered. This is presented as his own observation and recommendation, not a definitive, universally applicable technical flaw. It's his journey, not a scientific paper.
Original quote: “I find that these versions often get into repeated loops on the thinking and
just get stuck in that repeated loop.
so honestly, probably at the moment, I wouldn't bother with
using the abliterated ones.”
Unsloth model with MTP achieves around 120 tokens/second. — No Frame (75/100)
At 14:02
Another direct report of his test results. He's just laying out his numbers. 🔥
Why this score: The speaker is providing a specific performance metric from his testing of the Unsloth model. This is a direct, verifiable claim based on his setup, similar to his earlier benchmarks.
Original quote: “And this had really good, token rates.
with the MTP, I was getting around 120 tokens per second here.”
SGLang focuses on Blackwell GPUs and he's configured it for RTX Pro 6000, NVFP4, and D Spark. — Missing Context (45/100)
At 14:22
He says SGLang focuses on Blackwell, then shows it running on an RTX Pro 6000. Those aren't Blackwell. 🚩
Why this score: The speaker explicitly states SGLang focuses on Blackwell GPUs, which are NVIDIA's next-gen architecture. However, he then immediately shows his configuration for an RTX Pro 6000, NVFP4, and D Spark. The RTX 6000 is a professional Ada Lovelace GPU, not a Blackwell. This creates a disconnect between his general statement and his specific example, implying a broader compatibility or a different focus than stated.
Original quote: “so they are not supporting, all the different hardware.
They're really focused a lot more on the Blackwell, GPUs, and you can see here
that I've got it configured for doing the RTX Pro 6000, the NVFP4, and the D Spark.”
References an X account for proof — without showing it. Trust me, bro. 💀 — Anonymous Authority (45/100)
At 14:44
Says 'look at their X account' but doesn't show the account or the post. Just expects you to take his word for it.
Why this score: He's citing an external source for a claim about performance, but doesn't actually display the source. It's a classic 'go look it up yourself' move, which is fine, but it means the immediate evidence isn't there.
Original quote: “and if you look at their X account, they're kind of showing”
Claims 'much faster' tokens per second with 'their model weights' — still no numbers. 🙄 — Confidence Mismatch (45/100)
At 14:57
He's confident about 'much faster' but gives no specific numbers to back it up.
Why this score: The speaker uses vague comparative language ('much faster') without providing specific quantitative data to support the claim. This creates a sense of improved performance based on his confidence, but lacks the concrete evidence needed for objective evaluation.
Original quote: “but with their model weights, sure enough, I'm getting much faster tokens per second.”
Claims 'often' gets 200-220 tokens/second for 'not too long' things. Vague. 🤷♀️ — Confidence Mismatch (45/100)
At 15:08
He's using vague qualifiers like 'often' and 'not too long' for a specific performance claim.
Why this score: The speaker is making a performance claim (200-220 tokens/second) but qualifies it with subjective terms like 'often' and 'not too long.' This makes the claim difficult to reproduce or verify, as the conditions for achieving that speed are not precisely defined.
Original quote: “and you can see that for things that are not too long like this,
I can often actually get, around 200 to 220 tokens a second.”
Walks back the previous claim, says average is 'a bit under 200.' Classic volume game. 🙄 — Volume Game (45/100)
At 15:22
He just claimed 200-220, now he's saying the average is 'a bit under 200.' Contradicting himself in one breath.
Why this score: The speaker initially claims speeds of 200-220 tokens/second, then immediately follows up by stating the average is 'a bit under 200.' This rapid shift in numbers, without clear distinction or explanation, dilutes the impact of the higher claim and makes the overall performance unclear.
Original quote: “Now, I would say the average is probably a bit under 200 tokens per
second , maybe a little bit below that.”
Calls it 'flying' speed, but just said it's 'a bit under 200.' Hyperbole much? 😈 — Loaded Language (45/100)
At 15:30
He's using emotionally charged language like 'flying' after just downplaying the average speed.
Why this score: The speaker uses an emotionally loaded term ('flying') to describe the speed, despite just having stated that the average is 'a bit under 200 tokens per second.' This attempts to create a perception of exceptional performance that might not align with the actual numbers he's presented.
Original quote: “But at that speed that's flying I've got it running in a Docker
container, which is using their particular image from their cookbook.”
Declares Qwen3.8-27B the 'model to beat' for local AI, based on personal experience. — Confidence Mismatch (45/100)
At 16:47
He's calling it 'the model to beat' based on his own testing. That's a bold claim for a personal take. 💀
Why this score: The speaker presents a strong, definitive statement about the model's superiority ('the model to beat') as a universal truth, but immediately qualifies it with 'For me currently.' This is a personal opinion dressed up as a general recommendation, lacking broader comparative data or benchmarks beyond his own experience.
Original quote: “For me currently, if you can serve this model, this is the”
Uses an unnamed 'artificial analysis intelligence test' to confirm personal feelings about the model. — Anonymous Authority (45/100)
At 16:56
He cites an 'artificial analysis intelligence test' without naming it. That's not data, that's just a vibe with a fancy name. 🔥
Why this score: The speaker references an 'artificial analysis intelligence test' to support his claims about the model's performance, but provides no specific name, source, or link to this test. This makes it impossible to verify the test's methodology, results, or relevance, relying instead on an unnamed, unverified authority to bolster a personal opinion.
Original quote: “The fact that the artificial analysis intelligence test shows it to be so high just confirms for me a lot of things I felt using it and testing it out.”
Expresses optimistic future predictions and hopes for the model's improvement. — No Frame (75/100)
At 17:05
Just some hopeful speculation about future improvements. Nothing tricky here, just wishful thinking. 😈
Why this score: The speaker is openly speculating about potential future developments and improvements for the model ('I think it's gonna get even better,' 'I'm really looking forward to see'). This is presented as personal anticipation and not as a factual claim or a certainty, making it a straightforward expression of opinion and hope.
Original quote: “And I'd finish up by saying that I think it's gonna get even better. I'm really looking forward to see do we get a ThinkingCap version of this where it uses less tokens for the same amount of intelligence? Do we get other fine tunes that actually improve the intelligence?”
Declares the model a 'huge step forward' for local AI on 'prosumer hardware.' — Confidence Mismatch (45/100)
At 17:41
He says 'clearly a huge step forward' for 'prosumer hardware,' but the comments show people struggling with high-end gear. That's a disconnect. 💀
Why this score: The speaker confidently states the model is a 'huge step forward' for 'prosumer hardware.' However, community comments indicate that even with high-end consumer or professional-grade hardware (like RTX 6000 with 96GB VRAM or RTX 3090), users are getting modest token-per-second rates, suggesting that 'prosumer' might be a stretch for optimal performance, or that the 'huge step' isn't universally accessible as implied.
Original quote: “But clearly this is taking a huge step forward, both not only in open weights, but in local AI where you can run these things on sort of prosumer hardware”
See the full analysis with timestamps →