OpenAI released GPT-6.2 Astra, its updated frontier model, and says it now does a better job with slides, spreadsheets, documents, and coding. The company is pitching cleaner output and tighter template-following, but the only evidence in hand is the company’s own description of the change.
The most practical shift is in office work. OpenAI says GPT-6.2 Astra can produce slides, spreadsheets, and documents that “read less like AI,” which suggests fewer awkward turns of phrase and less obvious model-generated structure for people who need polished drafts fast.
GPT-6.2 Astra template control
OpenAI says the model more closely follows templates, which matters for anyone building repeatable slide decks or spreadsheet layouts because the output should stay closer to the format the user asked for.
That claim also sets a limit on how far readers should trust the announcement. The company says the model reads less like AI, but no independent benchmark or example in the material shows how much the writing changed or how often the new version stays on template.
GPT-6.2 coding tasks
The same release also extends to coding, where OpenAI says GPT-6.2 Astra is better at websites, gaming, and 3D design.
For users, that points to a model aimed at more than text polish. It is being positioned for work that mixes structure, code, and visual output, which is a different test from simply generating a cleaner paragraph.
GPT-6.2 Sol update
OpenAI is also launching an update to GPT-6.2 Sol, which it describes as smarter and more efficient than its predecessor.
GPT-6.2 Sol sits below GPT-6.2 Astra in OpenAI’s lineup, so the update gives users another option inside the same family. The unresolved question is how much better GPT-6.2 Astra performs in measured tests or real-world use than the prior version.







