GPT-6 Astra Demonstrates Extended Task Persistence, Running Complex Work for Hours Without Interruption
OpenAI's GPT-6 Astra is showing remarkable ability to sustain complex reasoning and multi-step tasks over extended durations, addressing a longstanding limitation in AI agents. The model, released September 4, 2026, can work continuously on computer-use tasks for approximately 40 minutes per benchmark (OSWorld 2.0) at 72.6% accuracy—47% faster than its predecessor GPT-5.6 Sol. More strikingly, in extended testing, Astra completed a full playthrough of Pokémon Fire Red in 18 hours, demonstrating hours-long persistence on sequential decision-making.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.