As one of the OpenBEL community members, I agree with what Gary has said. BioPax is much more structured and descriptive in regards to capturing pathways. BEL is designed for more of an overall view of biology representing multi-scale qualitative biology from molecular interactions to associating organism phenotypes to molecular events. There are currently more academic resources in BioPAX than BEL, but we are working to address that. Ingenuity and Thomson Metacore are available in BEL, and several bioPharmas are using BEL.
To me, BEL is the Biological version of the Chemical Reaction Language. Something easy enough to write on the whiteboard, but computationally tractable for reasoning both with and without data. Given data to process with it (e.g. RNA or Protein expression data or Metabolomics data), you can create quantitative results by combining data with the qualitative BEL knowledge representation.
BEL was created in private and used that way for years, but we felt it was useful beyond Selventa and is a living standard. We are in the last few weeks of deploying BEL 2.0 and have proposals to work on for BEL 2+ around genomic feature association to molecular events and population level effects. We are trying to balance complexity, computational tractability, and expressivity. At some point, we need to create a way to interoperate between BioPAX, SMBL and BEL as they all have specific strengths and weaknesses as they are tools for different purposes.