Expected Performance of Run-3 HLT b-quark jet identification

The developement of new heavy flavour jet tagging algorithms has made large progress with the utilization of more complex neural networks during Run 2. These developements are now deployed at trigger level with the DeepJet model exchanging the DeepCSV and CSVv2 algorithms for b-tagging trigger paths. The network is trained on trigger level simulation while previous ones were not trained explicitly on trigger level reconstruction. This application yields a better b-jet identification efficiency while reducing the light-flavour misidentification rate.