How will it get the performance of a Neo4J,Titan et al. with this model? Does it index them specially?
Seems odd to ignore Gremlin and OpenCypher just to have a proprietary / "yet another standard" graph query language.
One major difference is that ArangoDB has full support of transactions across collections. Also while MongoDB now has limited graph functionality, ArangoDB is actually optimized for graph queries and can do arbitrary traversals and joins very efficiently.
The problem with Gremlin/OpenCypher is that they are primarily designed as graph query languages. ArangoDB takes a multi-model approach and many ArangoDB users focus on documents first and add in graph queries as an optimization step as their products mature. AQL is designed to be very easy to learn and mixes the document and graph paradigms seamlessly, embracing the multi-model nature of the database.
Personally I think that starting with a blank slate was the right choice compared to other multi-model databases that have found it necessary to extend a subset of a standard query language with their own proprietary additions to achieve the same goals.
We did a benchmark that shows you the performance regarding Neo4J. Before you shout that it made by us have a look, please. It's all open source. The data, the scripts and so on. Everything we use for it, we described in detail.
AQL is much more than a graph query language; therefore, Gremlin and OpenCypher aren't alternatives.