Benchmarks

Which model scores best on each eval task, and which runs fastest?