Comment by CSMastermind

Using Fable, pretty much every request hit some gate they had for no discernible reason. These provider-level rejections should be incorporated into benchmarks as 0s on the tasks since that's the experience you'll actually get using the model.