đź”´LIVE: ChatGPT 5.5 Pro greatest AI model ever?
TL;DR
Alex Finn’s core claim is that ChatGPT 5.5 Pro is the most underrated coding model right now — after spending a weekend using Codex exclusively, he says it has overtaken Claude Code as his top “vibe coding” stack, even though he’d been a Claude and Cursor loyalist for more than a year.
In live build tests, 5.5 Pro consistently impressed on complex 3D outputs — it generated a working solar system simulator with orbital data, moons, speed controls, and detailed planet views, then followed it with a playable GTA-style sandbox featuring vehicles, weapons, police sirens, and a day-night city scene.
Claude still wins one important category: front-end design — Alex repeatedly says Claude’s UI work is “unbeatable,” calling ChatGPT’s front-end output its “Achilles heel,” and recommends keeping both subscriptions if you can swing roughly $200-$300/month.
Alex’s practical workflow advice is very specific: build step by step, not with giant one-shot prompts — he uses Codex on 5.5 medium with fast mode for day-to-day work, switches to Pro for heavyweight reasoning, and says the image generator is the best workaround for weak UI generation because you can design first, then code from the mockup.
A surprise detour was Clicky, a voice-first ambient agent that watches your screen and acts on commands — Alex was impressed that it could build a 3D rocket sim from a spoken prompt, but backed off hard on trust, saying an always-watching closed-source agent is too much of a privacy tradeoff unless it’s local or open source.
The stream doubled as a manifesto on creator integrity and staying close to the tools you actually use — Alex says he turned down a $30,000 Hostinger sponsorship because he doesn’t use it, argues most big AI creators are sponsored, and insists long-term trust beats short-term checks.
The Breakdown
The Pitch: Why 5.5 Pro Deserves More Respect
Alex opens hot: ChatGPT 5.5 Pro is “professional level amazingness,” the only model that gives him genuinely novel answers instead of bland averages. His big complaint is also its biggest reason people ignore it — it can take forever to respond — but he thinks the quality gap makes the wait worth it.
Codex Replaces Claude Code as Alex’s New Favorite
He says this shift is a big deal because he’s been publicly obsessed with Cursor, then Claude Code, since August 2024. After using Codex with ChatGPT 5.5 all weekend and “didn’t touch Claude Code once,” he declares the baton has officially passed, while making clear he’s not being paid by OpenAI and hasn’t taken sponsorships from tools he doesn’t personally use.
The Stream Wanders, but the Thesis Stays the Same
Before getting into demos, Alex veers into retro gaming, showing off a new Game Boy Color pickup and a stack of cartridges like Pokemon Red, Pokemon Gold, NFL Blitz, and a Kobe Bryant basketball game he calls “the worst video game I’ve ever played in my entire life.” It’s very him: half AI evangelist, half caffeinated nostalgia machine, using the tangent to make a point that older games felt like art while modern games are “hyper optimized” for addiction and microtransactions.
Test 1: A 3D Solar System, Built Live
For the first real benchmark, he has ChatGPT 5.5 Pro generate a solar system simulation with planets, moons, orbit speeds, planet stats, and camera controls. The first pass had dark planets and a minor error, but after one follow-up fix it became a slick interactive model where he zooms into Earth, Saturn’s rings, and Jupiter’s moons while practically narrating like a kid in a planetarium.
Claude Gets a Fair Shot — and Mostly Keeps Up
Alex runs the same solar system concept through Claude Opus for a side-by-side comparison. Claude’s version had better lighting and smoother polish out of the gate, but it missed some details like visible moons around key planets and had noticeable flickering; Alex ends up giving the win to 5.5 Pro while still admitting he prefers Claude’s warmer, more human writing style.
His Actual Buying Advice: Don’t Pick One if You Can Avoid It
Even while saying “gun to head” he’d pick ChatGPT right now, Alex tells viewers the real answer is to carry both OpenAI and Anthropic. His rough recommendation is $100 for Claude and $100 to $200 for ChatGPT, because 5.5/Codex is now his top coding stack, but Claude remains the king for front-end design and visual polish.
Clicky Steals the Show, Then Raises Red Flags
Late in the stream he tests Clicky, a new voice-driven desktop agent that follows the cursor and can spawn agents from spoken commands. He’s genuinely wowed when it builds a 3D rocket sim from a single voice prompt and casually comments on what’s already on his screen, but his excitement quickly turns cautious when he realizes the tradeoff: a closed-source assistant that sees everything on your display all the time is, in his words, a massive privacy sacrifice.
The Killer Demo: A GTA-Style Sandbox from 5.5 Pro
The most memorable moment comes when 5.5 Pro produces a playable Grand Theft Auto-like scene with sprinting, jumping, carjacking, weapons, police sirens, explosions, and a city you can drive through. Claude’s version was also impressive, but Alex’s reaction says it all when the ChatGPT build lets him hijack a car, shoot from inside it, and trigger chaos live: “That was the best one-shot game I’ve ever seen from AI.”