Skip to content
InYourGeek
visiteur@inyourgeek — shell
↹ compléter↑↓ historique⏎ ouvrirhelp
FR

The Monday letter · the tech and security week in one email

Free, every Monday, via Brevo (EU), one-click unsubscribe · your data and rights

AI· 3 min read

GPT-6.2 Astra and Sol: OpenAI promises quality and a smaller bill

OpenAI has unveiled GPT-6.2 in two flavors: Astra, the frontier model, and Sol, the everyday workhorse. API pricing shifts along the way, which is worth a look before your next invoice lands.

Illustration of OpenAI's two GPT-6.2 models, Astra and Sol, shown side by side with their API prices

OpenAI has unveiled its new model family, GPT-6.2, which Next covers in a brief published this November 18. Two variants are on the menu: Astra, the frontier model, and Sol, the one for everyday work. The usual disclaimer applies: everything below comes from OpenAI’s announcement as reported by Next, our only source at this hour. The qualities on display are therefore the vendor’s promises, not independent measurements.

Sol, the model you’ll actually use

Sol is the variant most users will end up with. OpenAI pitches it as the best balance of speed and efficiency for writing, coding, pulling several sources together or putting a budget in order. Its prose is said to be more natural and more concise, particularly for technical and professional writing.

The most persuasive argument fits in a single benchmark. On LifeVal, which measures performance on “tasks that matter in everyday life”, Sol reportedly beats GPT-6 Astra at one seventh of the cost. So the everyday model would outscore a frontier model, provided, of course, that your everyday life looks a lot like the benchmark’s.

Astra, or the slide that finally stays on brand

Astra is aimed at the most demanding work, with the emphasis on the quality of the documents it produces. OpenAI leans hard on presentations: slides are supposed to respect the layout, colors and typography of the source material, whether that source is a screenshot or the original file. A frontier model whose headline feature is obeying the brand guidelines: if you have ever sat through a review meeting about the wrong shade of blue, you know exactly what is at stake.

The rest of the list is more familiar territory. The writing is meant to be more direct, with less filler and fewer generic conclusions. Coding abilities are billed as improved, and software use is said to be more precise, notably for 3D design and simulations. OpenAI also promises a model that is comfortable with spreadsheets and with complex edits inside an application.

What it costs in the API

This is the most verifiable part of the announcement, because it ends up on an invoice.

  • GPT-6.2 Sol: 2 dollars per million input tokens, 10 dollars per million output tokens, and 0.10 dollar per million cached tokens.
  • GPT-6.2 Astra: 8 dollars for input and 40 dollars for output per million tokens, which is 20% lower than before.

Astra therefore costs four times as much as Sol, on input and output alike. If the promises about Sol hold up in real use, the question of which requests truly deserve the frontier model is going to keep a few architecture meetings running past the hour.

Who gets it, and when

According to the November 18 brief, both models are available today for Plus, Pro, Enterprise and Edu subscribers, in ChatGPT Work and Codex, as well as in the API. The source says nothing about free accounts.

One last detail: GPT-6.2 Astra promises less filler and fewer generic conclusions. We’ll lead by example and skip ours.

Sources (1)

Written with AI assistance from the sources cited above, then reviewed and approved before publication by Sébastien Soulier.

The Monday letter

The tech and security week, every Monday in your inbox

What to patch first, data breaches, the articles with pictures. Free, no ads, one-click unsubscribe.

Free, every Monday, via Brevo (EU), one-click unsubscribe · your data and rights