Fable 5 leads discovery test; GPT-5.6 answers every task
The Gate benchmark tests whether AI can reason toward major scientific discoveries published after its . Fable 5 achieved the highest , but its earlier version answered only 45% of the tasks and its latest version answered 39%.
The latest version also performed slightly worse on answered tasks, apart from refusing more questions. The reported Fable 5 scores include only tasks that both versions answered.
improved on GPT-5.5 and surpassed , giving it the strongest result among models in a similar class and price range. It remained slightly behind Fable 5, but answered 100% of the tasks, which could make it a more practical everyday tool than the latest Fable 5.
Key points
- Gate tests whether AI can reason toward discoveries published after its .
- Fable 5 recorded the highest .
- Fable 5's task fell from 45% in the earlier version to 39% in the latest version.
- The Fable 5 scores cover only tasks answered by both versions.
- trailed Fable 5 slightly but answered every task.