When Dijkstra Wins

Dijkstra handles any non-negative weights: $0$, $1$, $5$, $1000$. It is faster than Bellman-Ford when weights are positive.

The moment you see a negative edge, Dijkstra breaks. It assumes visiting a node again will not improve the distance.

If all weights $\geq 0$ and you need single-source distances, pick Dijkstra. With a binary heap, it runs in $O((N + M) \log N)$ time. For dense graphs where $M \approx N^2$, that is $O(N^2 \log N)$. Still faster than Bellman-Ford's $O(NM)$.