30% more token use, but even by their benchmarks, don't appear to have any real big successes there, and some regressions. What's the point? It doesn't do any better on the suite of obedience/compliance tests I've written for 4.6, and in some tests, got worse, despite their claim there it is better. Anecdotally, it was gobbling so many tokens on even the simplest queries I immediately shut it off and went back to 4.5.

Why release this?