GPT-5 Reduces the Need to Choose Models in ChatGPT
In his One Useful Thing newsletter, Ethan Mollick explains what’s interesting about GPT-5:
In general, AIs that “think” before answering (called Reasoners) are the best at hard problems. The longer they think, the better the answer, but thinking costs money and takes time. So OpenAI previously made the default ChatGPT use fast, dumb models, hiding the good stuff from most users. A surprising number of people have never seen what AI can actually do because they’re stuck on GPT-4o, and don’t know which of the confusingly-named models are better.
GPT-5 does away with this by selecting models for you, automatically. GPT-5 is not one model as much as it is a switch that selects among multiple GPT-5 models of various sizes and abilities.
It’s nearly impossible (and largely unnecessary) to keep up with all the changes in the AI world, but OpenAI’s release of GPT-5, the latest model to power ChatGPT, seems worth acknowledging. Figuring out which chatbot you prefer is hard enough, but even after choosing, you have to decide which large language model to use for each conversation. ChatGPT has long defaulted to GPT-4o, which is fast and cheap but offers weaker answers than more expensive reasoning models like OpenAI o3, which can take 10–15 minutes to return an answer. With GPT-5, ChatGPT analyzes your prompt and decides on its own which of several models of various sizes and abilities to use.
That’s both good and bad. If you never ventured beyond GPT-4o, GPT-5 automatically picking one of its more capable models could lead to much better responses. I believe this will be true for many users. On the other hand, if you carefully selected your models, you might be frustrated if GPT-5 chooses differently in a specific case. To accommodate such users, OpenAI lets paying users manually choose between GPT-5 and the slower, more capable GPT-5 Thinking from the model pop-up menu at the top. Those who are pining for GPT-4o can turn on Show Legacy Models in ChatGPT’s settings to access it as well.
In theory, GPT-5’s automatic model selection is where we want things to go because it allows the underlying models to improve while reducing the apparent change for change’s sake that so often afflicts tech companies. If OpenAI can make it so that ChatGPT seems the same—we humans excel at detecting personality changes—while improving its performance and reasoning behind the scenes, that could help reduce the incidence of tech whiplash.
Lastly, an Apple connection: Apple told 9to5Mac that Siri will be able to access GPT-5 in OS 26, but not before. But if you want to talk to ChatGPT, do so directly in the ChatGPT app rather than playing Telephone with Siri.
I’ve been using GPT-5 for a few days now and the results seem good - and fast - but that’s purely anecdotal. I don’t see GPT-5 Thinking as an option but there is an option for the Legacy models.
I’ve been using it fairly heavily for some text based work and I’m yet to hit a free version limit. I had intended to get a paid subscription (I’d cancelled my previous sub a few months ago) but the GPT-4 version I asked was very confused about the pricing models and plans. I asked about the pricing of GPT-5 and it stated there was a free model with limitations, or a Pro version for $200 US.
I asked if these were the only two options and asked about a mid range plan similar to the previous GPT-4 $30 plan:
I said $200 was too expensive and asked if there were other options:
After a bit more back and forth talking about how $200 US was beyond what most people could afford it finally came back with this:
Then I said “You just told me there were only two plans; free or $200 a month Pro. Why are you now talking about a Plus plan which includes GPT-5 for $20-30?”
All this to say that GPT-5 is available for free or on a couple of paid plans. CLEARLY you still need to check the results (especially on GPT-4o which is what I was chatting with) to ensure the accuracy of the results.
I have subsequently paid for the Plus subscription because I do use it quite a biit and think it represents good value.
I’ve been on the paid $20/month Plus plan for several months. I haven’t seen it offered for $30, and hopefully the price won’t increase to that point. And like Adam shared, I have a similar dropdown in both the Mac app and in the web browser, but mine doesn’t offer the legacy models that his shows.
I should mention I’m in Australia so it charges me $30 AUD which is near enough to $20 USD. I haven’t tried the browser version but in the Mac app I have (Version 1.2025.210) I definitely don’t see the GPT-5 Thinking option.
I checked for updates when I started and there was nothing available but I’ve just checked again and updated to Version 1.2025.217 - still no Thinking option.
Forgive me for I have sinned…
I thought I had subscribed but I must have stopped before the payment. I have just completed the payment and the Thinking option is now in the App. I was only aware of this because I finally hit the Free limit. Given the amount I have used it today it’s fair to say many, many people will be just fine using the free version - it seems very generous.
I used o3 almost exclusively because the answers were far better than 4o, and I wanted to see the chain of thought. Bringing back 4o doesn’t make any sense to me. I don’t want 4o back, I want o3. GPT-5 Thinking seems to be exactly the same which would be fine, but the option to see chain of thought is completely gone now.
From what I’ve read, you may still have to choose a model in some places. I’ve seen screenshots of Cursor where you chose between different levels of reasoning (low, med, high) and different speeds (normal, fast). eg. gpt5-low-fast and you can still see the chain of thought.
I think Cursor and other services that use OpenAI’s models will likely always provide options for switching between models because cost is always a factor then. The goal with reducing the need to choose models was specific to ChatGPT itself.
OpenAI says it won’t be so cavalier about removing old models in the future.
Interesting stats from OpenAI on how many more people are using reasoning models now that GPT-5 is automatically picking the model to go with the prompt. Reasoning model use went from 7% to 24% for paid users and from 0% (because it wasn’t available) to 7% for free users.