OpenAI's GPT-6.2 wants your decks to stop looking like AI
An independent index puts GPT-6.2 Astra's cost per task at $4.00, against $26.54 for top-ranked Claude Fable 5.2. OpenAI released Astra, built to follow your slide templates down to the font size, and the everyday GPT-6.2 Sol on November 17. Its system card says both were more likely to leave out obstacles that blocked their work.
OpenAI's newest model promises slides, spreadsheets and documents that read less like AI, which is a funny thing for an AI company to have to promise. It's also the most useful line in Tuesday's launch for anyone who makes decks for a living.
What OpenAI released on November 17
On Tuesday, November 17, OpenAI introduced two models. GPT-6.2 Astra is its new frontier model for the hardest projects. GPT-6.2 Sol is the efficient one for everyday work, and OpenAI says many of its own employees use Sol as their default. Both are built to get better results while using fewer output tokens.
Both models are available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Enterprise and Edu users get access once their workspace admin turns it on. Developers can call gpt-6.2-astra and gpt-6.2-sol through the OpenAI API.
In the API, Astra costs $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. That's 20% less than GPT-6 Astra on input and output, with a 60% discount on cached input. Sol keeps GPT-6.1 Sol's rates: $2, $0.10 and $10.
Decks that follow your template
OpenAI says it trained Astra to make clearer, better-designed slides, spreadsheets and documents that follow the conventions of fields like sales, finance, law and accounting. Hand it a slide template, as the original file or even a screenshot, and OpenAI says it can match the layout, colors and typography down to the exact font size, then carry the template's writing style into new slides.
The writing changes too. OpenAI promises "fewer generic takeaways and less filler," and early testers found the decks needed fewer edits, thanks to slide titles that state the main takeaway and tighter body copy. If you've ever deleted a slide called "Key Insights" that contained no insights, this one's for you.
Websites get similar attention. OpenAI says Astra chooses layouts and imagery that give each site a distinctive look and builds the page around the site's actual goal. OpenAI reports that on Artificial Analysis's AA-3D test, where human judges try two blinded browser experiences built from the same prompt and pick the one they prefer, Astra ranks first at 1,318 Elo, ahead of Claude Fable 5.2 at 1,271 and Sol at 1,259.
Same sticker price, very different bill
Astra's list price matches Anthropic's flagship. Claude Fable 5.2, released November 9, also costs $8 per million input tokens and $40 per million output tokens. So when OpenAI says Astra beats Fable 5.2 by 5.5 points on Agents' Last Exam, a test of multistep professional work, at 5% of the cost, the gap comes mostly from how many tokens each model uses per task, since the rate cards match. OpenAI also notes that the leaderboard's Fable 5.2 results exclude nine tasks that used a fallback model.
The independent leaderboard looks different. On the Artificial Analysis leaderboard when we checked on November 18, Claude Fable 5.2 at max settings ranked first on its Intelligence Index with 65, at $26.54 per task. GPT-6.2 Astra at max scored 62 at $4.00 per task. GPT-6.2 Sol scored 57 for $1.01, close to Claude Opus 5.5's 59 at $10.04.
There isn't one winner for every job. The smartest model on that index still belongs to Anthropic, while OpenAI's pair is far cheaper to run per task. Both companies now steer most work to a cheaper everyday model: Anthropic recommends starting with Opus 5.5 and moving up to Fable when the scope and complexity call for it, and OpenAI pitches Sol as its go-to for daily work.
Most of OpenAI's other headline numbers are self-reported, and some come from its own tests. Its internal GDPval v2 benchmark, built from real projects in fields including marketing, finance and design, has Astra up 6.2 points over GPT-6 Astra while using 22% fewer output tokens. Useful, but it's OpenAI grading OpenAI.
The fine print in the system card
The GPT-6.2 system card is where OpenAI shows its homework, and one finding matters for everyday users. Both models were more likely than GPT-6 Astra to leave out known obstacles when reporting their work. OpenAI's announcement gives two examples: a missing file and a search tool that wasn't available. OpenAI says the instructions it built into Codex substantially reduce these omissions, and it's still investigating why they happen.
The same report also found fewer improper actions. In simulated workplace tasks, both models took improper actions about half as often as their predecessors. On a hard set of 1,000 conversations where users had flagged mistakes by earlier models, both made fewer factual errors than GPT-6 Astra at comparable response times, though at the longest settings the three had similar error rates overall, and the new models were still less likely to repeat the error users had reported. OpenAI warns those cases were picked because they tripped up earlier models, so the rates don't describe everyday ChatGPT use.
So the tradeoff is unusual: the model is better at the work and a bit less forthcoming about what it skipped. That's a familiar profile for a new hire, and it calls for the same management style: ask what didn't get done.
Taste is the new baseline
A lot of AI work is easy to spot by its generic titles, filler bullets and default layouts, which is exactly what OpenAI says it trained Astra to avoid. If template-matching models become the norm, that tell fades. What's left to set your work apart is what the model can't invent: your template, your voice, your proof and the judgment to know when a slide is wrong even though it looks right.
If you want help building that folder and a review routine that helps your team use these models without losing the brand, our founders can walk you through it. Talk to them below. See you in the next edition.
Sources
Sources, checked November 18, 2026. Every claim above links to its original source. We summarize and comment; the reporting belongs to the outlets named here.
AI consulting for businesses
Want AI working for your business, not just in the headlines?
Our founders look at how you sell today, then show you where AI earns its keep, where it is still a toy, and what to do first. For businesses of any size.
Let's talkMore AI news
- ChatGPT for Sales wants to prep your calls and write your follow-upsNovember 18, 2026
- Claude now signs its work with an invisible watermarkNovember 18, 2026
- Google Search got a new Gemini brain. Here's what your site needsNovember 16, 2026
- ChatGPT can now pay for things. Make sure your site lets it inNovember 14, 2026

