These graphs are based on the construction proposed by my supervisor Anuj Dawar in this paper:
https://arxiv.org/abs/1809.08154
It essentially takes a random 3-XOR Formula, converts it into a graph, and applies the multipede construction using the the Cai Fürer Immerman gadget.
The construction contains a random component, so graphs produced with the same parameters will be different and have varying solve times. The graphs I have attached in dreadnaut and DIMACS format are from a single pass, and I can produce more difficult graphs of that size by retrying.
I have also included the solve times, which I obtained from running the graphs on the Cambridge High Performance Cluster: