OpenAI Launches Discounted GPT-6.2 Astra and GPT-6.2 Sol

OpenAI launches GPT-6.2 Astra, priced at 80% of the rate of 6 Astra, as well as GPT-6.2 Sol, which remains at unchanged prices. The former is presented by the company as intended for challenging projects, while the latter is for versatile everyday uses. Initial tests come with reservations about the results and interactions with the user.
Theo Browne criticizes technical arbitration requests
In feedback sent to OpenAI prior to the launch, Theo Browne emphasizes that the model can coordinate numerous sub-agents during lengthy tasks, but criticizes it for subsequently asking the user about internal choices that are impossible to resolve. After sending the command “/goal make a new level for Outer Wilds,” he reports that the model became stuck after two hours, asking whether the critic r24 should save his clips in 24_g521.mp4 or in 24_p11.mp4.
Another test submitted both models, set to Ultra in Codex CLI, to a three.js animation prompt: Yonkers was to rise as a giant robot and then traverse Manhattan, Queens, and Brooklyn using OpenStreetMap data. According to the person who conducted this test, the transformations produced by Astra and Sol are deemed less convincing than those of Claude Fable 5.2 and Gemini 4 Carbon. Initial feedback is described as mixed, with Astra 6.2 scoring lower than Fable 5.2 on several benchmarks and criticism regarding a communication style that can sometimes be confusing.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Astra costs 8 dollars per million in input
For Astra, OpenAI charges 8 dollars per million tokens for input and 40 dollars for output. Claude Fable 5.2 applies the same rates, which also represent 80% of those of its predecessor. The previous versions 6 Astra and Fable 5 were both charged at 10 dollars for input and 50 dollars for output.
The caching rate for Astra drops to 0.40 dollars per million tokens, down from 1 dollar previously, which is presented as advantageous for long agent sessions. Sol maintains a price of 2 dollars for input and 10 dollars for output per million tokens, like 6.1 Sol and 6 Sol, which is half the rates of GPT-5.6 Sol, which were 4 dollars and 20 dollars.
LLM already supports the new models
The API endpoints are gpt-6.2-astra and gpt-6.2-sol. The LLM tool has supported these models since its version 0.38, which has already been released.
GPT-6.2 Astra has also been used at the Ultra level in Codex on desktop to create a video montage comparing the animations of the two models. The request was for the first 10 seconds, with Astra in the upper half and Sol in the lower half, all in a square format of 2000x2000 pixels. The labels “GPT-6.2 Astra” and “GPT-6.2 Sol” were to be added aesthetically to each half.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.