GPT-5 has received mixed reviews, with notable issues in its routing system causing inconsistent performance and user frustration, though its advanced coding abilities and reasoning at higher settings demonstrate significant incremental improvements. While not the anticipated breakthrough to AGI, ongoing fixes and enhancements suggest potential for more efficient and specialized AI applications in the future.
The much-anticipated GPT-5 was released recently, but reactions have been mixed, with some prominent voices like Gary Marcus expressing disappointment. Critics argue that GPT-5 is overhyped and not the breakthrough path to Artificial General Intelligence (AGI) that many hoped for. There is skepticism about the massive investments by figures like Elon Musk and others into AI data centers, as these may not necessarily bring us closer to true AGI. Many users have reported issues such as hallucinations, routing failures, and performance below expectations on benchmarks, leading to frustration and even subscription cancellations.
A key feature of GPT-5 is its model routing system, which directs user queries to different versions of the model based on complexity and cost. However, this routing system has been problematic, often sending users to cheaper, less capable models, which degrades the experience. OpenAI acknowledged this issue, stating that the auto-switcher was broken but would be fixed soon. When users bypass the router and use the top-tier GPT-5 Pro or max reasoning effort models, the results are significantly better, showcasing impressive reasoning and coding capabilities that surpass previous models.
GPT-5 excels particularly in coding tasks, where it can quickly generate and iterate on complex software projects, such as games and 3D simulations. The model demonstrates strong instruction-following abilities and can create bespoke software tools efficiently, often outperforming its verbal reasoning in tasks that can be translated into code. This makes GPT-5 highly valuable for developers and users needing rapid prototyping and problem-solving through code generation, even if it struggles with straightforward math or some other tasks when relying solely on language-based reasoning.
Despite the criticisms, there is a consensus that GPT-5 represents an incremental improvement rather than a revolutionary leap. The model shows promise in specific domains, especially in coding and medium-horizon tasks that would take humans hours to complete. However, some feel that AI progress may be plateauing, with diminishing returns from simply scaling up model size. OpenAI’s team, including Sam Altman, has indicated ongoing improvements, particularly in fixing routing issues and making the system more transparent about which model is responding, which should enhance user experience moving forward.
In conclusion, GPT-5 is a mixed bag: it is not the AGI breakthrough many hoped for, but it is a powerful tool with notable strengths, especially in coding and complex task execution. The current issues with routing and inconsistent performance have led to divided opinions, but improvements are underway. Users are encouraged to test the model at its highest settings to fully appreciate its capabilities. The future of AI progress remains open, with potential for growth beyond just bigger models, focusing on smarter, more efficient applications.