Looking at their benchmarks, it's actually a better model than OpenAI and Jev (which in the same evals is worst than Jev?).