# GPT-6 is great. It's not AGI > A reviewer spent hours with GPT-6, hit the weekly usage ceiling in an hour, and still found a normal AI model underneath the hype. - Title: GPT-6 is great. It's not AGI · PCMAG | Shortify - Summary: A reviewer spent hours with GPT-6, hit the weekly usage ceiling in an hour, and still found a normal AI model underneath the hype. GPT-6 is the best tool on… - Keywords: gpt-6, openai, agi, ai coding, ai hype, technology, great, It's, PCMAG - Source: PCMAG — https://pcmag.com/opinions/i-blew-through-my-weekly-gpt-6-limit-in-an-hour-and-its-still-not-agi - Author: Ruben Circelli - Published: 2026-09-14T16:03:31+00:00 - Read time: 3 min - Topics: gpt-6, openai, agi, ai coding, ai hype, technology ## AGI means whatever sells The term stretches from beating humans at tasks to actual consciousness. Every AI company points at it as the goal, which makes the word useful for marketing and useless for measuring. > Artificial general intelligence (AGI) is a vague term that can mean anything from performance that surpasses human capabilities to true consciousness. ## The executives called it Nvidia's Jensen Huang said GPT-6 puts us in the AGI era. OpenAI president Greg Brockman agreed, which is not a shock coming from the company selling it. ## Hours of use, different verdict The writer spent many hours inside GPT-6 and calls the AGI framing a gross exaggeration. Amazing things, yes. A new kind of mind, no. > Yes, GPT 6 can do some amazing things, but here's what you should really expect from OpenAI's latest model. ## Benchmarks are mostly noise He has argued before that AI benchmarks are functionally useless to most people because scores never reconcile with real work. He still reads them, just not as truth. ## The ARC-AGI-3 jump On a test of how agents learn unfamiliar interactive tasks, GPT-5.6 scores 7.8% and Opus 5 scores 30.2%. OpenAI puts GPT-6 at 99.9%, a gap far bigger than a normal release. ## Coding feels different While building a Crimson Desert mod, GPT-6 flagged that GPT-5.6's planned mod format would not work and pushed him toward ASI format instead. That one worked. ## Dumbest setting beats the old best OpenAI's engineering lead says GPT-6 at its lowest intelligence setting outperforms GPT-5.6 at its highest. That explains why the difference is so easy to feel. ## Pro finally feels fast The Pro variant answers chat prompts much faster than GPT-5.6's Pro model, so he now reaches for it on questions that need deep thinking or heavy web searching. ## Computer use got wild It beats Pokémon faster than GPT-5.6 and finishes something as tangled as Rimworld. People are building Unreal Engine worlds with it and turning sketches into Blender models. ## It still needs babysitting GPT-6 introduces bugs, misses obvious fixes, and makes plain mistakes. On that same mod, it took hours of prompting at max intelligence to find a game-breaking bug it had created. > When coding, GPT-6 (like all other AI models) requires babysitting. ## An hour of usage, gone Crank the ultra intelligence setting, spin up a few agents, and a week of allowance on the $100 Pro 5x plan can vanish in an hour or two. He's done it more than once. ## Efficiency does not cancel the bill Max power is unnecessary most of the time and dialing it back stretches usage. Still, the efficiency gains do not outrun the heavier costs, especially on the $20 Plus plan. ## As a chatbot, a wash Asking about a Warframe update or cold brew containers, GPT-6 gave good answers. So did GPT-5.6. The chat experience barely moved, just like 5.6 over 5.5. ## His bar for AGI No double-checking outputs, no prompt engineering, no careful model selection or endless monitoring. Describe what you want, get it in short order. Consciousness not required. ## Real AGI would not be a quote If a company gets there, it outclasses the internet and the smartphone as the biggest advance in human history. It arrives as an earthquake, not as an executive soundbite. > If (or, perhaps, when) some AI company achieves AGI, it will be the single greatest technological advancement in the history of the human race. ## Try it, keep expectations sane Worth testing on your codebase or a 3D model. OpenAI reportedly has an internal model beyond GPT-6, used on the Navier-Stokes problem, and he doubts that one is AGI either. > GPT-6 is simply an improved version of GPT-5.6, nothing more or less. ## Key takeaway GPT-6 is the best tool on the market right now and an iterative one. Judge it by what it does for your work, not by what executives call it.