Is GPT-6 Astra Actually AGI? I Tested It Inside My JARVIS
Credibility score: 45/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "Is GPT-6 Astra Actually AGI? I Tested It Inside My JARVIS" and rated it 45/100 for credibility (a BS score of 55/100 — mixed credibility), on 2026-09-10. Its weakest claim — "GPT-6 Astra released + built full Jarvis from it" — scored 45/100 and was flagged as confidence mismatch. 13 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 13 claims analyzed: 0 scored under 40, 13 between 40 and 69, and 0 at 70 or above.
Claims analyzed
Teaser montage: GPT-6 called AGI, then roasted on iPhone crease — Just Vibes (50/100)
At 0:00
Opens with a highlight reel that sells the title premise before the real test begins.
Why this score: Merges the entire 30-second intro montage into one segment as instructed — no individual claims, just preview clips. The AGI question is teased, the iPhone joke lands, nothing to score on evidence yet.
Original quote: “[0:00] Jarvis, switch your brain to the GPT6 [0:02] Astra. [0:05] >> GPT6 Astra online. Sir, if I start [0:07] solving physics between your sentences, [0:09] do try to keep up. [0:10] >> Jarvis, I'm planning to get the new [0:12] foldable iPhone. What do you think? [0:14] >> $2,000 for a phone…”
GPT-6 Astra released + built full Jarvis from it — Confidence Mismatch (45/100)
At 0:48
Calls it 'GP6 Astra' like it's already out — the actual name is GPT-6 Astra, and it dropped a week ago, not yesterday.
Why this score: Speaker compresses the timeline to sound like he's on the bleeding edge the second it lands. Slight name slip plus 'just got released' gives the impression of instant access when the model is only days old.
Original quote: “So, GP6 Astra just got released and I used it to recreate Jarvis and give it brand new functionality.”
Tested GPT-6 vs others, 'absolutely remarkable' — Anonymous Authority (45/100)
At 1:49
Says he's tested it against the field — never names the other models, benchmarks, or any numbers.
Why this score: Classic empty comparison. 'Tested versus others' sounds authoritative until you notice there are no baselines, no scores, no receipts. Just the word 'remarkable' doing the heavy lifting.
Original quote: “I've been actually testing the GPT6 versus other models and honestly it's absolutely remarkable of how good this model is”
GPT-6 tool use 'very efficient' compared to rivals — Anonymous Authority (45/100)
At 2:06
Calls tool use 'very efficient' — again no numbers, no model names, just the vibe.
Why this score: Efficiency claim without a single metric. The sentence structure implies a head-to-head he never actually shows.
Original quote: “again the tool usage is also very efficient when it comes to GPT6 versus the other models there”
GPT-6 Astra search is 'absolutely amazing' — no comparison shown — Confidence Mismatch (45/100)
At 3:59
Calls it 'absolutely amazing' while naming no benchmark — just feels faster. 😈
Why this score: He's selling efficiency like it's settled fact. I've watched mortals hype every new model the same way since GPT-2. The moment lacks any data — just the word 'amazing' doing the heavy lifting.
Original quote: “the tool usage and the in the search, the internet search and all of the functionalities with GPT6 Astra is absolutely amazing how efficient it is”
Claims model did real research, not made-up info — Confidence Mismatch (45/100)
At 4:42
Says it 'did the research' like the model has a browser — still just pattern-matching web text.
Why this score: The speaker treats the model's ability to regurgitate pricing details as proof of genuine investigation. No source links, timestamps, or verification steps shown. The confidence that this is 'research' rather than sophisticated recall exceeds what's demonstrated.
Original quote: “it says that uh right now an alternative, the cheaper alternative is Kai uh Kie Kie. charges a few dollars a month with no subscription at all. Right? So, it gives you actually a researchbased information or it gives you all of the details based on the research that it did. Right? So, it's not just…”
Lists extra vision features as proof of advanced capability — Missing Context (45/100)
At 5:32
Slouching detection and phone alerts are useful scripts, not steps toward AGI.
Why this score: The speaker bundles several narrow computer-vision rules into one bucket labeled 'new functionalities,' implying emergent generality. These are pre-programmed triggers; the framing skips that distinction.
Original quote: “So, now Jarvis can actually see me through these eyes or the eyes functionality or feature that it has, which basically means that through the camera I can interact with them, right? So, and I can add multiple. So this eyes feature has a lot more functionalities like for instance if he sees that…”
Calls GPT-6 "remarkable" tool use — no benchmark given — Confidence Mismatch (45/100)
At 8:06
"Remarkable" without numbers — confidence doing the heavy lifting 🔥
Why this score: Speaker labels the tool performance "remarkable" yet supplies zero metrics or comparisons. The word carries the weight while the evidence stays home.
Original quote: “with the GPT6, the tool usage is also remarkable”
Calls tool access 'true AI assistant' — skips how narrow the tools actually are — Plain Sales Pitch (45/100)
At 8:30
Says 'just like a true AI assistant' after listing Telegram + Drive — that's a demo, not AGI.
Why this score: The speaker equates basic API connections with general intelligence. Tool access proves nothing about understanding; it's just permissions. Calling it 'true AI assistant' is marketing, not measurement.
Original quote: “more versatile depending on what we're using here or what your day-to-day um app usage is. Right? So if you're a designer or in finance or whatever it may be, this is where you can now give your personal assistant AI or personal AI assistant Jarvis those particular tools.”
Claims memory of brand = AGI — it's just saved files — Confidence Mismatch (45/100)
At 9:54
File access is not memory, mortal. It's a shared drive with better autocomplete.
Why this score: The speaker treats stored documents as if the model now 'knows' the brand. It doesn't. It retrieves. The leap from 'can read your old invoice' to 'true assistant that understands your business' is wishful branding dressed as breakthrough.
Original quote: “therefore, it has access to the formatting or your brand uh the way you do things like create invoices or whatever it may be. So that way you don't have to tell just like a true AI assistant, you don't have to give them all of the instructions and details every time, right?”
'Includes everything' — the usual upsell promise with nothing shown. — Plain Sales Pitch (45/100)
At 11:03
Promises 'everything' in a free pack, then immediately sells the 'finished version' that took months. Classic funnel. 💀
Why this score: The free pack is positioned as complete, yet the paid tier is the only way to avoid 'several months' of work. That contradiction is the entire sales mechanism, and the transcript offers no evidence the free version actually delivers.
Original quote: “this prompt pack includes everything, all of the details you need, including the uh different prompts that you can add these different functionalities”
Calls GPT-6 Astra 'really great' for building Jarvis — no benchmark cited. — Anonymous Authority (45/100)
At 11:27
'Really great model' with zero numbers or sources. Anonymous authority wearing marketing pants. 😈
Why this score: Speaker praises the model while naming no benchmarks, no ARC Prize results, no comparison to GPT-5.6 Sol. The praise stands on vibe alone, the same move OpenAI used when they said 'generational leap' without defining AGI.
Original quote: “this is starting to become really cool and I'm actually going to do more research and add more functionalities to Jarvis because like I said, this is a uh really great model that you can build a lot of things with it”
5-minute install claim — no demo, just a paid link. — Plain Sales Pitch (45/100)
At 11:57
Five minutes to a finished Jarvis, yet the only evidence is a zip file behind a paywall. Bold. Stupid, but bold. 😈
Why this score: Speaker asserts a five-minute path to the product he spent months building, without showing the clock or the steps. The claim exists only to justify the paid link, not because any timing data was provided.
Original quote: “if you just want like a 5minute install, uh again, this link is also going to be in the description”
See the full analysis with timestamps →