9 ms·

The amount of tokens/s is worrying: have they plateaud on actual reasoning ability and are now optimising for throughput?