Cisco open-sourcing vuln research models — feels like a trust exercise

I don't know if this is genuinely useful or just a way to get the community to do their homework for them. But I'm curious how well these models actually perform on real-world vulnerabilities vs the ones they've been trained on.