Since Mythos 5, Anthropic already went through 3 more internal flagship checkpoints. Probably the Mythos 6 pretrain with an increasing amount of RL.
Anthropic told METR that Model 2 was a new base model and they expect the later models from that base to be way bigger jumps over Mythos 5. Model 4 already hits >50% on every single benchmark they threw at it and staff self-report ~12x speedup.
They're sitting on it.