Perplexity’s CEO posted an important update on Reddit that answers a lot of questions about the current state of Perplexity after the past few weeks. I am recapping everything you need to know from that post. I will also share what I am seeing and what you can expect next.
Perplexity Mode Selector Changes: Auto mode by default
Auto mode is the new default. The idea is to let the AI decide if your question should be handled by regular, pro, reasoning, or deep research. That can make answers arrive as fast as possible.
As someone who uses Perplexity for work, I prefer to pick the mode manually. Different models shine on different tasks, and I want control. Even with auto as the default, you can still select the model you prefer.
Perplexity Mode Selector Changes: Model categories and missing selector
People asked why models feel inconsistent across modes and why the old selector is gone from Settings. Perplexity updated the UI, so instead of setting a default on the Settings page, you now pick the AI model when you start a new thread. That is why some users think they cannot find options like R1, Sonnet reasoning, or o3 mini.
Models are now grouped by type. Traditional chat models sit under one category with options like Sonar, Claude 3.7 Sonnet, GPT-4o, Gemini 2 Flash, and Grok 2. Reasoning models live under the Reasoning menu, where you will see R1, o3 mini, and Claude 3.7 Sonnet reasoning, and Deep Research is its own mode that currently uses Sonar.
I think this makes sense. There is no point in using reasoning for a quick answer. The main issue is communication, and users just need a clear cue that changing mode reveals a different model list.
Read More: Midjourney User Profiles
Perplexity Mode Selector Changes: How the selector works
If you want to choose a model, pick the mode first. Auto mode does not expose a model picker. Pro mode does, with choices like Sonar, Claude, and GPT.
Step 1: Start a new thread and open the mode switcher. Pick Pro, Reasoning, or Deep Research to see the models under that category. Auto will keep control on Perplexity’s side.
Step 2: In Pro, select the model you want for general chat. In Reasoning, pick options like R1 or o3 mini when you need step-by-step logic. In Deep Research, let the agent dig across sources for longer.
Step 3: If a follow-up flips you back to Auto, switch modes again. That is a small inconvenience, but you still have control. I do this regularly when I need consistency across a thread.
Step 4: I also hope Perplexity exposes image models soon. GPT-4o can already create images, and having a direct image generation option inside Perplexity would fill a real gap.
Read More: Hide Midjourney Images On Your Profile
Deep research and infrastructure updates
Perplexity is pushing Deep Research to think much longer, even 30 minutes or more. That should make the quality higher for complex tasks. It will take time, but the goal is better answers, not just faster ones.
They are also working on code execution with a live preview pane. Think of generating a site and seeing the preview update beside the chat. Tools like Claude Artifacts show how useful that can be, and bringing it into Perplexity would be a big improvement.
They are rewriting major parts of the system in Go. The older stack was not scaling well for what they want to ship next. The plan is to move carefully so new features land without adding more instability.
Read More: Hide Bulk Download In Midjourney
Follow-ups reverting to Auto
People noticed that Deep Research and Reasoning revert to Auto for follow-ups. That is connected to the auto-first approach. You can still switch back to your preferred mode at any time.
I see it as a minor inconvenience. Some users care a lot about that change, but manual control is still there. I switch it back when I want strict mode consistency across a thread.
Why no GPT-4.5
Perplexity enabled GPT-4.5 fast when it launched and even gave 10 daily credits, then reduced it to five, and now it is gone. The official line is speed: GPT-4.5 was too slow, around 11 tokens per second, while GPT-4o and Sonar output more. Removing it on speed grounds is the stated reason.
I do not fully buy that. GPT-4.5 is very expensive right now, something like 150 dollars per 1 million tokens, while GPT-4o is closer to 1 to 2 dollars per 1 million tokens. Even with limited credits, that would burn a lot of money, so I think cost is the real driver.
Read More: Midjourney Style Creator
UI bugs and the new answer mode
There are UI bugs and elements that seem to appear and disappear. That is tied to the new UI and the push toward an answer mode that feels more like Google. You search for things like the best restaurant or hotel and get a navigable answer view.
As they add more features, bugs are expected. The company is iterating on the interface and testing how the new answer mode fits. It will keep changing as they refine it.
Funding, pressure, and IPO timing
There are questions about funding pressure or a push to IPO. Perplexity says finances are solid and there is no attempt to squeeze users. The focus is on product quality and clearer communication to avoid bad assumptions.
They also say there will be no IPO until 2028. Recent changes are not prep for a near-term IPO. The priority is building and stabilizing the product first.
Final thoughts
Auto mode as the default makes sense for speed, but I still pick modes and models manually for control. Model categories are clearer once you get used to switching modes first, and the selector is right where you need it inside each mode. Longer Deep Research runs, code execution with previews, and a Go rewrite point to bigger features ahead.
Follow-ups defaulting to Auto is a small annoyance you can work around. GPT-4.5 is gone, officially for speed, but cost likely tipped the scales. Expect more UI changes as the answer mode matures, and do not expect an IPO before 2028.
