Is it useful or a flop? GPT-5 practical testing experience summary: a community debate with both "positive reviews" and "complaints" coexisting.

OpenAI has launched GPT-5, sparking a community debate filled with both praise and criticism over the past few days. This article dissects the initial user feedback. (Background: A comprehensive summary of OpenAI's strongest GPT-5 launch event: free access, key features, gpt-5, gpt-5-mini, and gpt-5-nano API pricing) (Context: OpenAI's open source weight model GPT-OSS: How does Apache 2.0 crown developers? CoT thinking chains unlock security sovereignty) This Friday (8th) morning, OpenAI officially introduced its latest model GPT-5, simultaneously releasing sub-models such as "thinking", "mini", and "nano", fully open for both free and paid users. Automatic model selection: convenience or black box? In recent days, community feedback has pointed out that some believe the model has become smarter, while others complain that GPT-5, under the automatic selection of new models, has been assigned to a less efficient version, with many users posting cases of "Nano generating incorrect answers". Currently, OpenAI's new mechanism has canceled manual model selection, requiring users to accept the system's judgment. For general users, this reduces menu anxiety; however, advanced users believe that it is difficult to determine whether the response is from GPT-5 thinking or Nano, making it hard to adjust prompting strategies. Senior AI observers have pointed out that this "one-click usage" simplified design is friendly to the public; after all, if every simple question is answered by the most powerful model, it does not align with cost-effectiveness and may slow down response speed; however, it may weaken the sense of control for professional users. Pricing and permissions: The outcry of Plus users GPT-5's pricing has also sparked discussions. Some users complain that the Plus plan has "severely weakened", and only the $200 Pro subscription enjoys full features. They angrily criticized: "ChatGPT has become worse for Plus users, the automatic selection almost only sends you to lower-tier models." However, there are also users who maintain a positive evaluation, with one user deliberately introducing a small error when throwing a corporate financial report to GPT-5 for analysis, and as a result, all models in the GPT-5 series could catch the contradiction; in contrast, GPT-4.1 and the old Mini mostly determined "no logical error". Additionally, some have summarized that GPT-5 is "more willing to provide clear decision recommendations" compared to its predecessor, reducing the avoidance of answers that hedge both sides. As for the tone of the responses, some believe that GPT-5 is more neutral, making it more suitable as an assistant; but some prefer the more flattering tone of 4o, which may be a matter of personal preference. Altman steps in Regarding online criticism that GPT-5 is not superior and a preference for comments about 4o, Sam Altman responded yesterday that they will reopen the use of 4o for Plus users and made the following points: GPT-5 launch updates: We will double the GPT-5 usage limits for ChatGPT Plus users as we complete the rollout. Plus users can choose to continue using 4o. We will observe usage to assess how long to continue offering legacy models. Starting today, GPT-5 will appear smarter. Yesterday, the automatic switch was down for a while, causing GPT-5's performance to seem less capable. Additionally, we are adjusting the decision boundary judgment method to ensure you receive the correct model more often. We will make the information regarding response models more transparent, so you can clearly know which model is answering your questions. The UI interface will update, making it easier to manually trigger the depth thinking mode. The overall rollout progress is slightly slower than expected. This is a large-scale change; for example, in the past 24 hours, our API traffic has nearly doubled. We will continue to work on stabilizing the system and will keep listening to user feedback. As previously mentioned, we did anticipate some bumps when pushing so many things at once, but... this time has been a bit bumpier than we expected! GPT-5 rollout updates: *We are going to double GPT-5 rate limits for ChatGPT Plus users as we finish rollout. *We will let Plus users choose to continue to use 4o. We will watch usage as we think about how long to offer legacy models for. *GPT-5 will seem smarter starting… — Sam Altman (@sama) August 8, 2025 Next observation Currently, GPT-5 has been online for only three days, and users need more time to test. Whether OpenAI can balance quality stability, transparent model allocation, and reasonable costs will determine whether GPT-5 can truly hold above the next competitive stage. Related reports Is your TikTok Shop still safe? AI + fake apps + phishing pages, the new type of scam "ClickTok" is targeting your crypto Wallet How long has it been since you seriously thought? Princeton scholars warn: AI is eroding the "legacy of the Enlightenment" Lawyer Lin Shanglun's special article》AI defeats three lawyers? Don’t get it wrong, this is the prologue of "Lawyer 2.0" <Is it useful or a dud? GPT-5 real-world testing experience summary: "Praise" and "criticism" coexist in community debate> This article was first published in BlockTempo, the most influential Blockchain news media.