When Anthropic's freshly released Claude Fable 5 became unavailable after a U.S. government directive, Microsoft Research principal Dimitris Papailiopoulos switched his chat work to Opus 4.6 and his autoresearch tasks to Codex 5.5 xhigh, noting that the prior Opus model still leads its 4.x family.
Kinda interesting to notice the natural pull toward models/setups, when the new best model is missing. In the absence of Fable, I switched back to Opus 4.6 in the chat, and for autoresearch full on Codex with 5.5 xhigh. I do think 4.6 has been the best of the 4.x Opuses.
Practitioners are openly ranking models by how well they handle sustained reasoning when the newest option disappears, turning routine downtime into an unplanned benchmark of staying power.
With Fable offline since June 12 and no confirmed restoration timeline, teams are testing whether last-generation models can carry demanding workloads without the expected leap in capability.
Many users reverted to Opus 4.6 after finding newer models like Fable lackluster and disappointing for autoresearch tasks.
@DimitrisPapail well well well
i am trying to work on the closest thing possible to a true "big model smell" eval which is to say: something that measures something that clever post training can't trivially gap, and is cheap + topically diverse i can't test mythos for obvious reasons, but... hmm...
@xeophon Well look at that I’m an expensive smell predictor
@DimitrisPapail Agreed, and same, Dimitris, everything past 4.6 has been lack luster for autoresearch.
I set a loop going to sleep last night on an optimizatino and when i got back tocolmp thjis morning I see my vast servers up and running and a message from opus 4.8 "im going to be honest, pushign further is impossiboe so im going to stop and save you money" - I canot fucking hate that model more, and the people who would program that level of retardation
@DimitrisPapail the way i also have a go-to model hierarchy when the newest one lets me down
opus 4.6 was really the ceiling for that gen