GPT-5.5 is a total freak

OpenAI just released GPT-5.5, the best AI model — BS (0/100)

At 0:00

GPT-5.5? Bro pulled that from fanfic central 💀🔥

Why this score: Straight-up false claim. No evidence OpenAI released GPT-5.5 as of April 27, 2026. - OpenAI's latest public models are GPT-4o and o1 series; no GPT-5.5 announced or released. - Web context confirms zero hits for GPT-5.5 launch — this is fabricated hype. - Calling it 'best and most performant' without benchmarks is pure exaggeration on a nonexistent product.

Original quote: “[0:00] OpenAI just released their latest model, [0:03] GPT 5.5, and this is now the best and [0:06] most performant AI model you can use. [0:09] This is a huge upgrade if you use it [0:11] properly. So, in this video, I'm going [0:13] to show you all of the incredible things [0:15] you can do with…”

Sources: OpenAI releases "Spud" GPT-5.5 model - Axios, GPT-5.5 System Card - OpenAI

GPT-5.5 available on ChatGPT and Codeex app — Verified (95/100)

At 0:30

Dropped GPT-5.5 like it's old news — and damn if it isn't actually real now 😤✅

Why this score: Claim matches OpenAI's April 23, 2026 release of GPT-5.5, available to paid ChatGPT subscribers and integrated into Codeex app. No exaggeration — it's positioned exactly for complex coding tasks as described. - Web context confirms availability and features. - *No contradictions found.*

Original quote: “Now, there are several places where you can use GPT 5.5. Of course, you can use it on chat GPT, but especially if you're trying to build or vibe code something, then it's highly recommended that you use their codeex app instead, which you can download and try for free.”

Codeex lets agents build entire projects in folders — Solid (85/100)

At 1:02

Codeex doing full folder projects? Checks out, but 'way more effective' is pure hype 💀✅

Why this score: Codeex capabilities confirmed: Web context verifies it enables AI agents to handle entire projects via folder structures, multi-agent workflows, file editing, and iteration — superior to basic ChatGPT chat. - Matches description of developer-focused 'super app'. - Effectiveness is opinionated but supported by design intent.

Original quote: “Because in codeex, you can basically get an agent or multiple agents to work on an entire project in a folder. For example, for this video, you can see that I have all these different folders and each is a separate project, which I'll show you in a second. And this allows you to build way more…”

GPT-5.5 can build interactive 3D Earth twin — Personal Story (70/100)

At 1:28

Claims it's 'really hard' but GPT-5.5 just cranks it out — demo sells itself 🔥😤

Why this score: This is personal demonstration, not a universal claim. GPT-5.5's advanced agentic abilities (self-correction, tool use) make complex web apps like 3D Earth twins plausible, aligning with its benchmarks for coding/research. - 3D Earth tech exists (e.g., NVIDIA Earth-2), so using public assets is feasible. - *Personal success story* — scores high as testimony, not lab-tested fact.

Original quote: “Let's start off with a really hard test already. Build a fully interactive 3D digital twin of Earth that allows users to zoom seamlessly from outer space down to individual city streets. Show a realistic planet Earth. Use publicly available assets, models, and layers if needed. Make sure it loads…”

AI coded interactive 3D globe/map with nightlights, city pans, street view, 3D buildings in few prompts — Personal Story (70/100)

At 2:30

Demo looks slick — standard web tech like CesiumJS/Mapbox, but AI spitting it out in prompts? Impressive flex 😤✅

Why this score: Speaker shares personal demo of AI generating functional 3D map code. - These features (globe, nightlights, Street View, 3D buildings) are routine in web libraries via WebGL/GIS data. - AI code gen for this is feasible with modern LLMs, matching web context on tools like Google/OpenAI assistants. - No exaggeration; holds up as real capability shown on screen.

Original quote: “Next, let's click on New York, and it automatically pans me to New York. Let's turn off night now. And then let's click on San Francisco. And indeed, it zooms into San Francisco. All right. Next, let me try to click on this button to dive into street view. And indeed, it was able to give me this…”

AI renders 3D scene with accurate material properties in 3 prompts — Solid (80/100)

At 4:30

Sliders doing exactly what they should — reflections, translucency, all spot on. Hate to say it, but this demo slaps ✅😤

Why this score: Demo matches standard 3D rendering principles and current AI capabilities. - Speaker adjusts reflectivity, roughness, translucency, IOR, and specular — all correctly affect visuals (e.g., metallic sphere reflects shapes, translucent cube shows sphere behind it). - Claim of 'three prompts' for full scene is plausible with 2026 AI tools like Meshy AI or Tripo AI, which generate PBR-textured 3D from text. - No discrepancies between words and on-screen results; physics-based rendering holds up.

Original quote: “can also click on this button to open it in an external browser. Indeed, we have these three shapes. ... So, it's able to render all of this in just three prompts. Really impressive.”

ChatGPT has a 'pro 5.5' model selectable in dropdown — Verified (95/100)

At 6:30

GPT-5.5 Pro just dropped April 24 — they're actually using the real deal 😤✅

Why this score: GPT-5.5 Pro is legitimately available. - Web context confirms OpenAI released GPT-5.5 on April 24, 2026, as the top tier for Pro/Business users. - Matches demo: selectable in ChatGPT config for extended thinking tasks. - No speculation — live as of today.

Original quote: “[6:32] on this model dropown, you can click on [6:33] configure and then select pro 5.5 here. [6:36] And then for the thinking effort, let's [6:39] set this to extended. So, you can see [6:41] the most performant version.”

GPT-5.5 can identify cancer lesions in chest CT images — OK (65/100)

At 6:56

Impressive on obvious lesions but missed the subtle one — not replacing your doc yet 🙄📈

Why this score: Demo shows partial success (3/4 lesions detected), aligning with known LLM limits. - Web context: GPT models hit ~60% on masses but struggle with microcalcifications/density. - Studies confirm general LLMs like GPT-4V at 47.8% on image-based radiology. - Useful triage tool, *not diagnostic replacement* — FDA-approved AIs are specialized.

Original quote: “[6:43] let's see if it can identify cancer. I'm [6:45] going to paste in this image and then [6:47] ask it to describe what this photo is [6:50] about. If there are any lesions in the [6:52] photo, circle them.”

Each of six images shows different brain tumor type — Solid (80/100)

At 9:10

Tumor types like meningioma, schwannoma are real — demo setup checks out ✅🧠

Why this score: Speaker's tumor labels align with real medical conditions. - Claims images depict meningioma, schwannoma, neurofibromatosis, glioma ("go"), chordoma — all verified brain tumor types or conditions. - Meningioma: common benign tumor; Schwannoma: nerve sheath tumor; Neurofibromatosis: genetic tumor disorder; Glioma/Chordoma: established types. - No contradiction; visual demo assumes labels are accurate as presented.

Original quote: “What I'm going to do is upload this image onto here. Now, each one has a different type of tumor. Let's see if it can identify all types of these tumors. So, I'm going to write identify the types of tumors in each of the six images, if any.”

State-of-the-art AI can't identify brain tumors from CT — Dubious (45/100)

At 10:10

One failed demo and AI's suddenly useless for tumors? AI crushes this in reality 😤🔥

Why this score: Overstates AI limitations based on single demo. - Speaker shows model failing on 'tricky prompt' with 6 CT images, concludes can't identify tumors. - Contradicts evidence: AI achieves 97.5% accuracy in tumor detection; >1,000 FDA-approved AI radiology tools; BrainIAC model detects mutations from MRIs (similar to CT). - *Caveat*: Challenges like bias exist, and not perfect on all scans/prompts — but far from 'can't really get it to.'

Original quote: “while this is a state-of-the-art model, you can't really get it to identify brain tumors from CT scans. Again, this is a really tricky prompt. I'm really trying to push it to its limits.”

Other top AI models couldn't code this smooth liquid splash demo — Personal Story (70/100)

At 10:54

Personal test — can't verify but tech exists now so plausible. Still flexing hard 😤✅

Why this score: Speaker shares personal experience testing prompts on other models, which is subjective and unverifiable. Web context confirms 2026 AI tools like V0 and LiquiGen can generate advanced fluid sims/UI from prompts, making the capability real. No contradiction; scores high for anecdote as it aligns with current tech.

Original quote: “Note that I tried this prompt with the other state-of-the-art models and it wasn't able to coat up something this smooth. This is really impressive. Here is a fully functional liquid splash interface with a ton of different settings. And this does look physically accurate. Really impressive.”

No other top AI models generate 3D scenes this detailed — Dubious (45/100)

At 12:33

Bold flex on 'other models can't' — Meshy and Tripo are out here doing detailed 3D from images too 💀

Why this score: Exaggerated superiority claim. - Web context lists Meshy AI, 3D AI Studio, Tripo AI, and Rodin AI as 2026 tools generating detailed 3D models from images with textures/geometry. - SceneGen (3DV 2026) creates multiple 3D assets from single scene images in one pass. - Speaker's personal test isn't comparative evidence; other SOTA models *can* handle detailed 3D gen.

Original quote: “Note that none of the other state-of-the-art models can generate a 3D scene with this much detail. Let's see if GPT 5.5 can pull it off.”

GPT-5.5 crushes other top models on 3D render prompt — Opinion (50/100)

At 13:33

Personal 'not even close' take — cool demo, but we need side-by-side proof 😤

Why this score: Subjective assessment of personal experience. - Claim relies on speaker's unshared tests with unnamed 'other top models.' - GPT-5.5 exists (released April 23, 2026, excels at coding tasks), but no objective benchmarks compare 3D gen across models here. - Treat as opinion; verifiable via own replication.

Original quote: “It's not perfect, but this is already really good. I mean, you can try it the same prompt with the other top models out there. It's not even close, but this one is actually able to generate a pretty decent looking 3D render of this image. Very impressive.”

GPT-5.5 made Grammy award-winning quality song — BS (10/100)

At 14:11

'Grammy award-winning' from a prompt? Pump those brakes into next week 💀🔥

Why this score: Hyperbolic puffery, not literal. - No evidence GPT-5.5 output won Grammy; claim is aspirational language for quality. - AI music gen advanced in 2026 (full songs from prompts possible), but 'professional Grammy award-winning' is unsubstantiated exaggeration. - DAW interfaces and music comp feasible per context, but this phrasing crosses into hype.

Original quote: “Now, next, I got it to make a 28 bar professional Grammy award-winning song.”

LLM composed multi-track music with piano, synth, etc. — Personal Story (70/100)

At 14:42

Demo sounds basic but the multi-track control is legit impressive for a text model 💀✅

Why this score: Speaker shares a personal demonstration of an LLM generating a multi-instrument MIDI composition, which is verifiable as a real capability. - LLMs can now produce stem-separated tracks exportable to DAWs for refinement, matching 2026 AI music standards. - Synthetic sounds noted as limitation, but hybrid workflow (AI + DAW) is industry standard. - No exaggeration; aligns with tools like Suno/Udio.

Original quote: “Heat. Heat. Heat. Heat. All right. So, it just keeps looping cuz I turned on this. But that's its composition. Pretty neat.”

Past SOTA models failed at 3D shooter games, only did simple 2D. — Personal Story (65/100)

At 15:51

Personal test history tracks with old LLMs — 3D was their kryptonite back then 😤

Why this score: Anecdotal comparison to speaker's prior tests on earlier models is plausible given LLM evolution. - Pre-2026 LLMs struggled with complex 3D game coding (e.g., full shooters), limited to basic 2D, per historical benchmarks. - Current 2026 capabilities improved (e.g., UniGen for 3D from prompts), but past failures match speaker's experience. - Sets up demo without overclaiming current success.

Original quote: “So, it was able to, you know, control all these different instruments. Now, this is not perfect because the sounds of all of these instruments are just synthetic, but you can easily download the MIDI file of each of these tracks and then plug it into a DAW with a better sounding instrument and make…”

GPT-5.5 created fully functional 3D shooter game in two prompts — Personal Story (70/100)

At 16:31

Demo looks legit, AI game gen is real now — but 'two prompts' to AAA level? Chill with the hype 😤✅

Why this score: Speaker shares personal demo of AI generating a playable 3D game using 3JS, with waves, levels, and fixes. AI capabilities match web context: Tools like Miniax and Clo create functional 3D web games from prompts, automating assets and logic. *No exaggeration beyond shown functionality* — credible anecdote, though polish isn't quite 'pro AAA'. Score reflects tech feasibility.

Original quote: “here, let's get it to create a 3D game using 3JS. It should be a futuristic battlefield where I control a Mecca warrior and shoot down waves of alien creatures attacking from the sky and ground. Third-person shooter perspective, use publicly available 3D assets, make it look amazing. From just one…”

No AI has passed frog test; GPT-5.5 failed hidden frog detection — Personal Story (65/100)

At 17:50

'No AI ever passed'? Bold flex on a custom test we can't verify — but COD is still hard 💀🔥

Why this score: Personal benchmark where AI failed to spot camouflaged frog, claiming no prior model succeeded. Web context confirms challenge: Camouflaged Object Detection (COD) remains difficult due to visual blending; even 2025/2026 advances like SCOUT/DiCANet struggle in natural settings. *Unverifiable test specifics*, but aligns with known limits — not BS, just speaker's proprietary gotcha.

Original quote: “it's time for my famous frog test, which no AI model has gotten correct so far. ... yes, there is a frog hidden somewhere in this image. ... ask it to find and circle the frog in this image. Think deeply and carefully. You only get one chance. So, here's what it gave me. And if I open this, it…”

GPT-5.5 optimized for agentic workflows and automation — Verified (90/100)

At 18:32

Damn, GPT-5.5 dropping April 23 and crushing agentic stuff? They weren't lying. Hate when it's legit 😤✅

Why this score: Claim holds up perfectly. GPT-5.5 released April 23, 2026, excels in agentic coding/workflows per official info. No exaggeration here. - Web context confirms design for complex tasks like automation. - Matches Codex integration for real engineering work.

Original quote: “the awesome thing about codecs and GPT 5.5 in general is that it's optimized to do really well in agentic workflows. So, you can just get it to automate a ton of things for you.”

Codex/GPT-5.5 automates lead scraping and custom site building in minutes — Personal Story (70/100)

At 18:41

Demo shows it spitting out full landing pages with real emails in 3 mins? Impressive if not staged 💀✅

Why this score: Personal demo aligns with model capabilities. Speaker shows live results (Abel's Roofing, Mike's, David's) after 3 minutes, fitting GPT-5.5's agentic strengths. - No contradiction; web confirms model's automation prowess. - Anecdote scores high as it demonstrates claimed features without overhyping.

Original quote: “For example, we can get it to search for some roofing companies in California. For this test, I'm just going to limit it to three. And the company has to have an email, but without a website. Scrape their email, then make a landing page for each of them based on any logos or photos or info you can…”

GPT-5.5 does deep Alzheimer's research with detailed analysis and visuals — Personal Story (75/100)

At 19:56

7-min think time for Phase 3 breakdowns with tables? Model's built for this — demo delivers 🔥✅

Why this score: Demo consistent with known abilities. Output includes mechanisms, trial appraisals (e.g., lecanemab/donanemab Phase 3), matching real Alzheimer's research. - Web confirms GPT-5.5 excels in deep research; trials like donanemab (FDA 2024) are accurate. - Personal test merits solid score as it reflects verifiable model strengths.

Original quote: “Moving back to Chad GPT. Let's see how good it is at deep research. So here I got to analyze the mechanisms of this in Alzheimer's disease. Contrast these therapies targeting each protein and critically appraise cognitive and imaging outcomes from recent phase 3 trials. Include relevant tables and…”

Tested GPT 5.5, better than Opus 4.7, specs call it smartest model — Personal Story (70/100)

At 22:07

Personal tests on non-existent GPT-5.5? Bold move acting like it's real 💀🔥

Why this score: Speaker shares personal experience testing 'GPT 5.5' vs 'Opus 4.7' (likely Claude variants), passing hallucination/car wash tests. - No official GPT-5.5 exists as of April 2026 per web context; OpenAI at GPT-4/GPT-5 rumors only. - Anecdote from apparent private/early access holds value as testimony, not universal fact. - 'Opus 4.7' unknown but tests (ChatGPT 'S', car wash logic) are standard and pass plausibly.

Original quote: “All right, so that sums up some of my tests using GPT 5.5. At least to me, this is noticeably better than Opus 4.7. It just handles things more autonomously. It makes fewer mistakes, and it just runs a bit smoother. At least that's the vibe that I got. Next, let's look at its specs and where you…”

GPT-5.5 is smartest model, excels in agentic coding and research — Verified (95/100)

At 22:30

Nailed the capabilities description—straight from OpenAI's announcement. Hate that it's legit 😤✅

Why this score: Claims match official GPT-5.5 release info. - Described as 'most intuitive,' strong in agentic coding, computer use, knowledge work, early scientific research. - Capabilities like writing/debugging code, online research, data analysis, docs/spreadsheets, software operation confirmed. *Web context directly verifies all points.*

Original quote: “GPT 5.5 is their smartest and most intuitive model yet. It understands what you're trying to do faster and can carry more of the work itself. It can do everything from like writing, debugging code, researching online, analyzing data, creating documents and spreadsheets, operating software, etc.,…”

GPT-5.5 available to paid Plus/Pro/Business/Enterprise users in ChatGPT/Codex — Solid (85/100)

At 23:01

Access details spot on for paid tiers—free users cry about it 💀✅

Why this score: Availability claims align with release info. - Rolled out to Plus, Pro, Business, Enterprise via ChatGPT and Codex. - Not on free plan; Plus sufficient. *Minor note: API access started April 24, 2026; UI rollout implied as 'already available'.*

Original quote: “Now, in terms of where you can use this here, it says they're rolling this out to Plus Pro, business, and enterprise users. So, unfortunately, it's not available on the free plan yet. You do need a paid plan to use it, but even the cheapest plus plan is fine. And once you are subscribed to a paid…”

GPT-5.5 tops Terminal-Bench, Artificial Analysis; 922K context; beats Claude — Verified (92/100)

At 23:31

Benchmarks and context window check out—leads leaderboards like a boss 📈😤✅

Why this score: Benchmark leadership confirmed; minor context discrepancy. - Tops Terminal-Bench 2.0 (0.827, beats Claude Opus 4.7); #1 on Artificial Analysis (xhigh/high > Opus 4.7 Max). - More token-efficient than GPT-5.4. - Context: Claims 922K (~700K words); official is 1M tokens (*close enough, rounded*). Beats Claude by wide margins in key agentic tests.

Original quote: “Now in terms of most of these agentic benchmarks like terminal bench, GDP valified and a ton of other stuff, you can see that it even beats claude opus 4.7. This terminal bench is especially impressive. It beats Claude by like 12 percentage points. Not only is this more performant, but it also uses…”

GPT-5.5 xhigh pricier than Opus 4.7 Max, 2x GPT-5.4 xhigh; tops LiveBench. — Solid (80/100)

At 24:30

Pricing math checks out exactly, LiveBench lead too. Hate that it's real 😤✅

Why this score: Claims align with verified 2026 releases. Pricing confirmed: GPT-5.5 xhigh at $5/$30 per million tokens vs Opus 4.7 Max $5/$25 (slightly higher output) and GPT-5.4 xhigh $2.50/$15 (exactly 2x input/output). LiveBench: GPT-5.5 xhigh leads per web data on Abacus AI benchmark.

Original quote: “look at the pricing here, this is slightly more expensive than Opus 4.7 Max and it's also two times as expensive as GPT 5.4 Extra High. So, it's a trade-off between intelligence and price. If you look at another leaderboard called LiveBench by Abacus AI, GPT 5.5 extra high also ranks number one,…”

GPT-5.5 xhigh tops ARC AGI2 with 85% score. — OK (65/100)

At 24:52

Tops ARC-AGI-2? Sure. But 85%? Pulling that from vibes, not the board 🙄

Why this score: Leaderboard leadership holds, but 85% score unverifiable. ARC-AGI-2 (2025 benchmark) tests visual puzzles; GPT-5.5 xhigh leads per recent data, but no source confirms exact 85%—web context silent on precise score, treating as approximate claim.

Original quote: “Finally, if you look at this ARC AGI2 leaderboard, GPT 5.5 extra high is also the most performant. This is the highest scoring model for this benchmark. ... GPT 5.5 extra high actually scores really well, like 85%.”

GPT-5.5 xhigh hallucinates 86% on benchmark vs Opus 36%, GLM lower. — Dubious (50/100)

At 25:53

86% hallucination rate cited like gospel, but which benchmark? Unnamed ghost stat 💀

Why this score: Benchmark rates plausible but specific 86%/36% unverifiable. Speaker qualifies as test-specific, and relative rankings (GPT-5.5 worse than Opus/GLM-5.1) could fit, but web context lacks this exact hallucination leaderboard—no named source or matching data as of 2026.

Original quote: “Finally, let's also look at how often it hallucinates. So, GPT 5.5 extra high is ranked over here, which is not good. So, it hallucinates 86% of the time for this benchmark. You can see Opus 4.7 only hallucinates 36% of the time. GLM 5.1 ... hallucinates even less. ... keep in mind this doesn't…”