OpenAI Launches GPT-6.2 Astra and Sol in ChatGPT Work, Claiming It Beats Claude Fable 5.2 at 5% of the Cost

OpenAI released GPT-6.2 Astra and Sol in ChatGPT Work on 17 November 2026. The business states Astra scored 5.5 portion points greater than Anthropic’s Claude Fable 5.2 on a professional-work criteria’s partial-credit step, at about 5% of Fable’s approximated expense per job.

The test, Agents’ Last Exam, determines how well a design finishes tasks that take numerous actions. OpenAI’s outcomes likewise put Astra 5.2 points ahead of its predecessor, GPT-6 Astra, at less than a 3rd of the earlier design’s expense.

OpenAI compared Astra at its ‘xhigh’ thinking setting with Fable 5.2 at its optimum setting. Astra’s rating covered all 122 jobs. Myth 5.2’s rating covered 113 after the leaderboard left out 9 jobs that utilized a fallback design.

Developers pay by the token, a system of text, for the product they send out to a design and for its reaction. OpenAI’s basic API rates for input and output on the brand-new Astra are both 20% lower than for GPT-6 Astra. Sol keeps GPT-6.1 Sol’s costs.

A Spreadsheet Test for GPT-6.2 Astra

OpenAI created Astra for tasks such as developing an assessment workbook or preparing a financial investment memo. Both tasks start with choosing which figures to draw from the source files. Synthetic Analysis’s AA‑AnalystAgent tests that ability on genuine spreadsheets and reports.

In one project, the design should determine California Medicaid’s costs on medical transportation from the program’s expense reports. Throughout the standard, Astra addressed 62.5% of the concerns properly on all 5 independent efforts, according to OpenAI.

For GDPval v2, its internal follow-up to GDPval, OpenAI made use of tasks in consulting, financing and style to check how its designs manage longer projects. Astra scored 6.2 portion points greater than its predecessor while producing 22% less output tokens. Sol enhanced on GPT-6.1 Sol by 5.3 points.

How GPT-6.2 Astra Fared in Law and Science

Each legal project on Harvey LAB‑AA v1.1 includes a comprehensive quick. A design needs to discover the pertinent realities throughout a number of files and use them in a piece of work that satisfies every requirement. A submission with a product accurate mistake, such as a developed citation, ratings no on the criteria’s primary pass-rate procedure.

OpenAI states Astra passed at more than two times the rate of Claude Fable 5.2 while performing at about a sixth of the expense.

Astra for Law, a variation set up for legal representatives, scored greater still. OpenAI likewise trained the basic design to follow the design templates experts supply for slides, spreadsheets and files. It explains Astra’s composing as ‘more direct and particular, with less generic takeaways and less filler’.

Where GPT-6.2 Sol Came Close to Astra

Sol is created for daily work. OpenAI states the brand-new variation can finish requiring tasks with less output tokens, permitting individuals on paid strategies to end up more jobs before reaching their use limitations. Much of the business’s staff members turn to Sol initially for both technical and non-technical work.

On LifeVal, OpenAI’s test of difficult daily jobs, Sol outscored GPT-6 Astra at about a seventh of the older design’s expense.

OpenAI states Sol scored within 0.5 portion points of Astra on Cognition’s FrontierCode 2.0 coding criteria at about one-fifth of Astra’s expense per job, with both designs checked at their optimum thinking settings.


Discover more from PMN S.P.O.R.T.S - A PRIME MEDIA NETWORK BRAND

Subscribe to get the latest posts sent to your email.

Related Articles

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Captcha verification failed!
CAPTCHA user score failed. Please contact us!

spot_imgspot_imgspot_imgspot_img