OpenAI touts GPT-6.2’s power—and its promise to use fewer tokens

OpenAI says its latest Astra and Sol models improve creative and coding work while using tokens efficiently. Independent benchmark figures shared by Perplexity suggest gains, but do not by themselves establish how the models perform across everyday tasks.
OpenAI touts GPT-6.2’s power—and its promise to use fewer tokens

OpenAI touts GPT-6.2’s power—and its promise to use fewer tokens
OpenAI has released GPT-6.2 Astra and Sol, presenting the pair as updates to its frontier and more efficient model lines. Coverage of the launch says Astra is designed to produce slides, spreadsheets and documents that “read less like AI” and follow templates more closely, with coding improvements in websites, games and 3D design. Sol, meanwhile, is described as “smarter and more efficient” than its predecessor.

At launch, OpenAI CEO Sam Altman emphasized speed and cost-conscious use, saying the models “can do amazing things and are very token-efficient.” He singled out Sol’s “ultrafast” mode as useful for everyday tasks. That pitch puts efficiency alongside capability: the company is selling not only stronger outputs but also a model it says can deliver value with fewer tokens.

A separate measure came from Perplexity, whose CEO Aravind Srinivas shared results on the company’s Wide-and-Deep-Research benchmark. Astra scored 0.783 at an estimated $12.54 per task, up 0.095 from GPT-6 Astra; Sol scored 0.746 at about $2.65, a 0.065 increase over GPT-6.1 Sol. Srinivas characterized the results as “Good improvements” in research performance. The figures offer a specific, task-based case for progress, though they are benchmark results—not a broad verdict on performance across every use.

Together, the launch claims and Perplexity’s numbers frame GPT-6.2 as an incremental upgrade with two selling points: more polished work and better efficiency. The practical test will be whether those gains hold up beyond the company’s descriptions and a single research benchmark.

https://foxvector.com

Write a comment