16 October 2026
Epoch tests six frontier AI models on real research tasks
First reported
Deep Learning Weekly ran this on .
- Epoch, an AI research organization, graded six leading AI models on tasks drawn from its own research work.
- The models were reliable on tasks with clearly defined answers.
- They struggled with open-ended parts of the work, where there was no single right answer.
Our read
Only Deep Learning Weekly among the 30 newsletters covered this Epoch evaluation. The archive's earlier stories on this thread are the OpenAI pay-per-completed-task test from 2026-09-01 and the 2026-08-18 story on AI systems designed to work together.
Earlier on this story
Written by Newskeryx from which newsletters covered this story and from our own archive.
How it was covered
Deep Learning WeeklyEditorial team
Evaluation of six frontier models on Epoch's own tasks, with results on well-defined versus open-ended work