I tested GPT-5.4 Thinking, and it gave me great answers (until I dove deeper) (www.zdnet.com)

OpenAI has released GPT-5.4 Thinking, a model designed for more complex and cognitively demanding tasks. Unlike its predecessors, this version focuses on "bigger thoughts and challenges," showing significant improvements in text-based responses and thoughtful analysis across various prompts. Available for use in the API and within paid ChatGPT plans, initial tests revealed that while the model offered insightful answers, especially on topics like social media and travel itineraries, it struggled with image generation and occasionally missed the mark on formatting and direct response to queries. This release is noteworthy for the AI and machine learning community as it demonstrates advancements in the model's reasoning capabilities, particularly in nuanced understanding and detailed analysis. Positive highlights included a well-structured critique of social media's role in society and comprehensive travel planning suggestions. However, issues arose with the AI's tendency to produce irrelevant answers or poorly formatted responses, emphasizing the need for ongoing supervision during interactions. Ultimately, while GPT-5.4 Thinking shows promise for tackling complex inquiries, it raises concerns about its reliability in adhering to specific instructions, signaling that further refinement is essential before broader deployment.