[vladimir_gor]

I'm a big fan of benchmarks and now finally we have one to evaluate models on physical tasks. Will be interesting to see how fast this gap will narrow.