OpenAI unveils GPT-6 Astra with computer-use skills, faster reasoning
OpenAI has launched GPT-6 Astra, a model that can interact directly with apps and spreadsheets, scoring 92.7% on an interface understanding test. Experts raise safety concerns about its new reasoning.
On the OSWorld 2.0 benchmark, which tests computer-use performance, GPT-6 scored 72.6 percent, compared with GPT-5.6's 65.7 percent and Claude Opus 5's 70.2 percent. OpenAI's latency simulations put the average task at roughly 40 minutes instead of 75 minutes, a near-50 percent reduction. Astra also scored 92.7 percent on ScreenSpot-Pro, which measures a model's ability to understand where to interact with interfaces; GPT-5.6 Sol scored 76.9 percent and Claude Fable 5 scored 87.3 percent.
The new model introduces a reasoning architecture called recurrent depth, which allows it to think about a problem repeatedly before producing an answer. OpenAI said this could improve accuracy in high-stakes enterprise situations. However, the company admitted that monitoring Astra's reasoning has been harder than monitoring Sol's chain-of-thought reasoning, raising safety concerns. As a result, Astra is being rolled out gradually and monitored for now.
Astra is available to Plus, Pro, Business and Enterprise users, with a Pro variant accessible to Pro and above. Free and Go subscribers are not getting access to the flagship model, according to TechRadar.