(Credit: Zooey Liao/PCMag Composite; GettyImages/ OpenAI)
ChatGPT's major new model releases have historically been huge news because they introduced innovative and surprising features and improvements. GPT-5 was supposed to do that, too: In fact, OpenAI CEO Sam Altman once compared GPT-5 to the Manhattan Project. Unfortunately, these seem to be empty words. GPT-5 has some upgrades I wanted based on my experience with competitors, but it's largely an insignificant update.
Unlike some, I’m not desperate to have OpenAI’s older models back, but GPT-5 has let me down enough that I’m considering a switch to Claude or Gemini for my AI chatbot needs. My opinion might change as I continue to use GPT-5, but below are my thoughts after trying it out for a few days.
GPT-4 Had Ample Room to Improve
ChatGPT’s 4-series models are great, which is why ChatGPT tops our list of AI chatbots. That said, all chatbots (including ChatGPT) are prone to giving incorrect responses and hallucinating. I wanted to see many fewer instances of both behaviors with GPT-5.
I also wanted OpenAI to increase ChatGPT’s context window so that it was more competitive with alternatives. I thought ChatGPT could also benefit from quality-of-life improvements to its interface, along with more advanced integrations like those in Claude and Gemini. Other chatbots also hold advantages when it comes to API pricing, interactive 3D characters, privacy policies, realistic video generation with audio, voice modes, and more. Essentially, I wanted GPT-5 to meaningfully bridge as many of these gaps as possible.
GPT-5 Feels Familiar, If Underwhelming
GPT-5 is supposed to perform better than GPT-4 by cutting down on deception and hallucinations, but I haven’t noticed a significant improvement in these areas. Just like with GPT-4, GPT-5 struggles to answer certain questions, confidently presenting incorrect information or just making stuff up. With both GPT-4 and GPT-5, I notice these issues most when asking niche, specific questions that require up-to-date information, like how Soulframe mechanics work as of the latest update or whether a certain RimWorld has been updated to support a particular version of the game. These issues don't seem to happen more often than they did with GPT-4, but they don't happen a lot less, either.
(Credit: OpenAI/PCMag)Based on OpenAI’s pricing page, which now includes details about GPT-5, context windows across the Free, Plus, and Pro tiers of ChatGPT are the same as they were with GPT-4 at 3K, 32K, and 128K, respectively. Meanwhile, Gemini has a 1M context window, and Claude has a 200K context window. GPT-5 still lags far behind the competition in this respect.
With GPT-5, Gmail and Google Calendar integrations are reportedly on the way. But it's difficult to care too much about these because Gemini already works seamlessly with Gmail and Google Calendar, alongside pretty much every other Google Workspace app. OpenAI might develop more integrations eventually, of course.
Finally, GPT-5 doesn't bring any meaningful changes to ChatGPT’s interface. A new model doesn’t necessarily guarantee user experience improvements, but I still would have liked to see at least some tweaks.
On the brighter side, GPT-5 does have some new features, such as the ability to create apps from scratch. Based on my initial testing, this feature works similarly to Claude’s equivalent: It’s fun to play around with as somebody who otherwise couldn’t write a single line of code, but it takes a good deal of prompting to get apps working.
API pricing is another win for GPT-5, which now compares more favorably with low-cost competitors such as DeepSeek. GPT-5’s cheapest model, GPT-5 Nano, costs $0.05 per input and $0.40 per output, which represents decent savings over its next-cheapest model, GPT-4.1 Nano ($0.10 per input and $0.40 per output). However, GPT-5 Nano is a brand-new model, so it's not yet clear whether it's worth the asking price for developers.
The voice chat mode also gets some upgrades. ChatGPT’s most advanced voice tech will soon be available for free users, while paid subscribers will get nearly unlimited access to it. You will also soon be able to choose among different personalities to customize GPT-5's voice chat mode. However, none of these changes make its voice chat more lifelike, which is what I most wanted to see with GPT-5.
Will I Unsubscribe From ChatGPT?
GPT-5 has, undoubtedly, introduced some of the upgrades I asked for, such as better API pricing, fresh integrations, new features, and voice chat improvements. However, GPT-5 doesn't deliver the things I wanted to see most, such as a bigger context window and more accurate responses. I’m not the only one who isn’t fully down with GPT-5: Many online are expressing dissatisfaction, to put it lightly.
My opinion might change as I spend more time with OpenAI’s latest model, however, so keep an eye out for my updated ChatGPT review. For now, I plan to keep using GPT-5 in my personal life. But if GPT-5 really is as similar to GPT-4 as it seems, and I can't find anything to take advantage of in my day-to-day usage soon, I won’t be paying for ChatGPT any longer.
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.