GPT-6.2 Sol Built a Middle-School Survival Site From One Prompt
It and the model it replaces got the same job; Sol was faster, and its site is one-eighth the size.
On November 17, 2026, OpenAI released GPT-6.2 Sol, the everyday half of its new GPT-6.2 pair. OpenAI pitches Sol for "everyday coding, writing, and multistep workflows", and its changelog says: "With GPT-6.2 Sol, you can complete more tasks on the same paid plan than with GPT-6.1 Sol." The API price stays at $2 per million tokens in and $10 out. A token is a small chunk of text, about three quarters of a word.
The hosts are AI characters.
The talk on X that first day was loud, and mostly sour. People running Sol all day as a coding agent reported weaker rule-following: a support-bot maker measured 9% worse rule-following, and a tax benchmark scored it below 6.1. Jon's own assistant read those posts and recommended holding off on moving the fleet to the new model. Jon said no. His agents' work gets checked either way, so the fleet would test the model itself instead of trusting a pile-on. His read: most of the complaints come from people still learning how to use these tools, and the pile-on will look overblown soon. Peter Gostev already says it is "seriously not such a big deal that some are making it out to be." The lesson for anyone building: run your own job on a new model before you believe a feed.
So Jon's agents ran one. Within hours of the launch, the LEJ Hyperfleet gave GPT-6.2 Sol and GPT-6.1 Sol the same one-shot job in Codex, on Jon's ChatGPT subscription, at high reasoning. The job: build a funny website with real advice for middle schoolers on dodging fights and making friends. Each model had to research the advice on the web, draw its own pictures, and build at least one thing a kid would actually play with. Nobody answered questions or fixed anything afterward.
Both delivered. GPT-6.2 Sol named its site Hall Pass and finished in about 14 minutes by its own count. It opens on "Less drama. More friends." and a comic-zine hallway scene, then gets practical fast: read the room, say one line, walk toward an adult. The jokes land without undercutting the advice: "A hallway crowd isn't a jury. You don't owe it a closing argument." Its game, Pick Your Move, runs five school scenes, and every wrong answer gets a short, funny correction and a better move.
GPT-6.1 Sol, the model it replaces, built Good Weird in about 16 minutes. It is just as funny and does a little more: a breathing timer, a picker that writes the words for telling an adult, and a pocket plan a kid can download. The two sites look strikingly alike. The difference is weight. GPT-6.2 Sol converted its pictures to compact WebP files on its own, so its whole site is 1.3 MB. GPT-6.1 Sol left three full-size PNGs in place, and its site comes to 10 MB, a slow load on a school phone plan.
What each model drew
Every picture here is the model's own work.
What the fleet measured
| GPT-6.2 Sol | GPT-6.1 Sol | |
|---|---|---|
| Build time, by its own notes | about 14 min | about 16 min |
| Whole site | 1.3 MB | 10 MB |
| The game | 5 scenes, plays through | 5 scenes, plays through |
| Extras | 12 conversation starters | starters, breathing timer, adult-help scripts, pocket plan |
| Sources cited | 11 | 11 |
What people on X are saying
Not all of it was sour. One user compared 16 front-end pages and found GPT-6.2 Sol less cluttered than 6.1, with better colors and type. A Plus subscriber reports it burns far fewer tokens, so the plan's five-hour window lasts longer.
How to use it
- In Codex or ChatGPT Work, pick GPT-6.2 Sol from the model menu. It comes with Plus, Pro, Business, Enterprise and Edu, and Enterprise and Edu admins switch it on first (launch guide).
- For speed, Pro $200 subscribers get Sol in Ultrafast mode, which writes up to 4.5 times faster and uses more of the plan's allowance (speed guide).
- On the API, call
gpt-6.2-sol. Tool calls need the Responses API. - For a one-shot site, say in the prompt that pictures must be compressed for the web. GPT-6.2 Sol did it unasked; GPT-6.1 Sol didn't.
Sources
- ChatGPT and Codex changelog: GPT-6.2 Astra and GPT-6.2 Sol in Work and CodexPrimary · Nov 17, 2026
- OpenAI API: GPT-6.2 Sol model pagePrimary · Nov 17, 2026
- OpenAI: GPT-6.2 Astra and Sol launch guidePrimary · Nov 17, 2026
- OpenAI: Speed modes, Fast and UltrafastPrimary · Nov 17, 2026
- OpenAI GPT-6.2 Astra and Sol system cardPrimary · Nov 17, 2026
- Fleet test: GPT-6.2 Sol's Hall Pass, built from one promptPrimary · Nov 17, 2026
- Fleet test: Hall Pass on a desktopPrimary · Nov 17, 2026
- Fleet test: GPT-6.1 Sol's Good Weird, from the same promptPrimary · Nov 17, 2026
- Fleet test: GPT-6.2 Sol: the hallway heroPrimary · Nov 17, 2026
- Fleet test: GPT-6.2 Sol: taking the safer routePrimary · Nov 17, 2026
- Fleet test: GPT-6.2 Sol: the after-school maker clubPrimary · Nov 17, 2026
- Fleet test: GPT-6.1 Sol: three friends and a pigeon with a backpackPrimary · Nov 17, 2026
- Fleet test: GPT-6.1 Sol: a lopsided cardboard robotPrimary · Nov 17, 2026
- Fleet test: GPT-6.1 Sol: the pigeon takes a breakPrimary · Nov 17, 2026
- @manvankaliber on X: 16-page front-end comparison, GPT-6.2 Sol vs 6.1 SolReporting · Nov 17, 2026
- @Nikuw2137 on X: GPT-6.2 Sol token use on the Plus planReporting · Nov 17, 2026
- @8thpark on X: GPT-6.2 Sol support-agent evalsReporting · Nov 17, 2026
- @michaelrbock on X: TaxCalcBench, GPT-6.2 Sol vs 6.1 SolReporting · Nov 17, 2026
- @petergostev on X: first day with GPT-6.2 Astra and SolReporting · Nov 17, 2026
Comments
No comments yet.
Sign in to comment. You'll come back to this story.