found an old thread where support confirmed o3 runs at “high” on perplexity, curious if same is true for claude since nobody agrees here - some say medium, some say low/minimal, one thread just died with no answer
also there’s a bug report where “high” effort throws a 400 error on claude opus 4.5 via api but low/medium work fine, so maybe high just isn’t wired up right for claude models
anyone actually know the real number, or is this just guesswork at this point
Hey @Slowianin ! Great question. The model effort level will default to the provider’s default level. Anthropic’s default is high for Sonnet 5 (Anthropic docs).
appreciate the answer! that clears up the effort question. but if high is the real default, why do so many people (myself included tbh) feel like claude on perplexity reasons less than on claude.ai directly?
found a couple things that might explain it - silent fallback to a cheaper model during high load, plus apparently a lower default temperature than claude.ai uses natively. any of that actually true on the backend, or is it just placebo from people misreading the model chip at the bottom of responses?
Hey! Interesting insights. If you have a request that looks off, please send the request ID to api@perplexity.ai. There isn’t silent fallback: if you request claude-sonnet-5, any retry/failover happens only between hosting providers serving the exact same model unless you explicitly pass a fallback chain yourself, and the response’s model field always reflects what actually ran. Like reasoning effort, Temperature is untouched: if you don’t set it, we send nothing and the model provider’s native default applies.
thanks, that clears up the fallback/temp stuff on the api side at least. my vibe comment was about the regular chat product though not the api, so not 100% sure how much of this carries over
if i had to guess now it’s less about hidden parameter tweaks and more just the wrapper itself, system prompt + whatever gets injected from the search/rag step before claude even sees my actual question. would explain the different feel even with identical effort/temp under the hood tbh
not trying to be dramatic about it but for reasoning/vibe i still just go to claude directly over the perplexity wrapper. mostly wanted to actually understand why instead of assuming it’s some dumbed-down-model conspiracy