- Microsoft researchers determine that current LLMs aren't good at long-running tasks
- More interactions and less structure significantly reduce benchmark performance
- "Python is the only domain where most models are ready"
New research from a trio of Microsoft workers has uncovered a fundamental issue that could be blocking effective agentic AI -namely that most AI models can't actually reliably handle long-running workflows.
To quantify their findings, the researchers introduced a new DELEGATE-52 benchmark to provide metrics across 52 sectors, including coding, accounting, science and more.
Ultimately, the paper concluded current LLMs "introduce sparse but severe errors that silently corrupt documents, compounding over long interaction."
AI isn't that good at long-running tasks, yet
The study goes into some of the latest AI models including Gemini 3.1 Pro, Claude 4.6 Opus and GPT-5.4. It found that even they "corrupt an average of 25% of document content by the end of long workflows," with lesser models even more likely to get things wrong.
The DELEGATE-52 benchmark uses real documents at around 15K tokens in length and introduced 5-10 complex editing tasks with a "round-trip relay simulation" that asks AI to perform a transformation then reverse it. This allows the researchers to measure how effectively each model reconstructs the documents back to their original forms.
Highly structured and programmatic areas were where the models performed best, with the Microsoft researchers concluding that "Python is the only domain where most models are ready." Conversely, natural language workflows, creative areas and semi-structured documents saw model models struggle.
The paper also uncovers that, the longer the token length, the more likely an AI model is to struggle.
Where frontier models differed was not in their ability to eliminate errors – just that they were able to delay errors. Some of the other models tested by Microsoft's researchers included a number of GPT-5 and GPT-4 generations, Claude options, Gemini models and one each from Mistral, xAI and Moonshot – totalling 19 different models from six families.
Gemini 3.1 Pro took first place with a DELEGATE-52 benchmark score of 80.9% after 20 interactions; Claude 4.6 Opus (73.1%) and GPT-5.4 (71.5%) round out the top three, and GPT 5 Nano (10.0%) falls into last place.
In short, the paper concludes that today's AI models are not reliable enough to be trusted for long-running, autonomous workflows, highlighting key areas where model developers must focus on in the future and offering up yet another benchmark to determine model capability.
Via The Register
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.