How long can an AI model work independently on a complex task before it needs human help?
Much longer than it previously did, according to the latest research, and it's accelerating faster than ever.
METR (Model Evaluation and Threat Research), an organization that tracks autonomous AI capabilities, just released its updated "Time Horizon 1.1" report. The findings: Claude Opus 4.5 can now autonomously complete tasks equivalent to 320 minutes of human work. GPT-5 reaches 214 minutes. Capabilities are doubling approximately every 89 days, which is about 20% faster than previously estimated.
To understand what this means for businesses and knowledge workers, I talked it through with SmarterX and Marketing AI Institute founder and CEO Paul Roetzer on Episode 195 of The Artificial Intelligence Show.
AI systems are getting dramatically better at sustained autonomous work, and the rate of improvement shows no signs of slowing.
"It was doubling every seven months,” says Roetzer. “Now it seems like it's doubling every six months, basically.”
He doesn’t anticipate a slowdown in the future.
METR's research has focused primarily on coding and coding-related tasks, where Claude has particularly excelled. But the underlying principle applies broadly to knowledge work.
Roetzer emphasizes the importance of personalizing these findings to your own work.
"When can you start thinking about something that takes you an hour or two hours or 10 hours to do as a human worker? And where can these models now do that work?" he says.
The threshold for meaningful automation isn't just capability but also reliability. And that's where things get complicated.
"As good as agents are becoming, in most industries they are still largely unreliable,” Roetzer says. “They still require humans in the loop, and they're nowhere near as autonomous as people think they are.”
But they’re getting better quickly, the METR report shows.
A chart in the report tracks this important trend: a line going up every time a new frontier model is released. This upward movement, showing the increased autonomy of AI, is critical to watch.
Because then the question becomes: When will AI begin to handle autonomous work in your field? And are you ready?