OpenAI empowers users: New GPT-5 modes arrive, GPT-4o returns.

OpenAI is recalibrating its approach to user interaction with its latest flagship model, GPT-5, by introducing a suite of manual controls that cede more authority to the user. In a significant shift from the automated, one-size-fits-all system that initially characterized the GPT-5 rollout, ChatGPT users can now select from three distinct operational modes: "Auto," "Fast," and "Thinking." This change, coupled with the notable reinstatement of the popular GPT-4o model for paid subscribers, signals a strategic pivot by the AI research lab, directly addressing user feedback and the growing demand for more granular control over the powerful, and at times personal, experience of interacting with artificial intelligence. The move suggests a maturation in the AI industry, where raw technical capability is increasingly being balanced with user experience, customizability, and the nuanced preferences of a diverse user base.

The newly introduced control system for GPT-5 is designed to offer users a tailored experience based on their immediate needs.[1][2] The default "Auto" mode is intended for the majority of users, functioning as an intelligent switcher that automatically decides whether a query requires a quick response or deeper, more complex reasoning.[1][3][4] For users who prioritize speed, the "Fast" mode delivers quicker, more immediate answers.[5][6] Conversely, the "Thinking" mode is engineered for tasks that demand more in-depth analysis and detailed reasoning.[1][5] To manage computational resources, OpenAI has placed a weekly rate limit of 3,000 messages on the "Thinking" mode; once this cap is reached, users can continue with additional capacity through a supplementary "GPT-5 Thinking mini" model.[1][3] Further enhancing its capabilities for complex tasks, the "Thinking" mode supports a substantial 196k-token context limit, allowing it to process and analyze much longer documents and conversations.[1][3] This tiered system of control allows users to consciously trade speed for depth, optimizing the AI's performance for specific applications ranging from rapid brainstorming to meticulous data synthesis.

This implementation of user choice marks a direct response to significant user backlash that followed the initial launch of GPT-5.[7][8] When GPT-5 was first rolled out, it completely replaced the existing model selector, forcing all users onto the new, auto-switching system and removing access to previous models, most notably GPT-4o.[7][9] A wave of criticism quickly emerged across social media, with many longtime users lamenting the change.[5][7] They described the new GPT-5's personality as "colder," "mechanical," and less conversational than its predecessor.[5][7] Some users speculated that the move was a cost-saving measure by OpenAI, designed to route queries to less computationally expensive models without user consent.[5][10] The outcry was swift and loud enough that OpenAI CEO Sam Altman announced a reversal, confirming the return of GPT-4o to the model picker for all paid subscribers.[1][7] Furthermore, a "Show additional models" toggle was introduced in the settings for paid users, unlocking access to other legacy models like o3 and 4.1, solidifying the retreat from a mandatory, unified system.[1][3][11]

The rationale behind this pivot extends beyond simply placating user complaints; it reflects a deeper learning process for OpenAI about the human-AI relationship. Sam Altman acknowledged the feedback, stating, "one learning for us from the past few days is we really just need to get to a world with more per-user customization of model personality."[5][3] This sentiment is being put into action, as OpenAI is reportedly working on an update to GPT-5's personality to make it "warmer," while aiming to avoid what some users found "annoying" in GPT-4o's more effusive style.[1][5][3] The episode highlights a fundamental tension in the design of user-facing AI: the push for seamless, automated efficiency versus the demand for transparency and control.[12][13] By providing explicit modes, OpenAI empowers users to become active participants in their interaction with the AI, choosing the right tool for the job rather than being subject to the decisions of a black-box router. This level of control is becoming a crucial aspect of the user experience, particularly for professionals who rely on the AI for specialized and complex tasks.

The implications of OpenAI's decision reverberate throughout the competitive AI industry. This move elevates the importance of user experience, interface design, and even AI "personality" as key differentiators, suggesting that the race for AI dominance will not be won on performance benchmarks alone.[14][8] Providing granular controls and respecting user preferences for specific model behaviors could become a new standard for competitors to meet. The decision to reserve the full range of model selection for paid subscribers also carves out a clear value proposition for premium tiers, framing flexibility and choice as features worth paying for.[8][15] Ultimately, this event serves as a powerful case study on the influence of a vocal user community. It demonstrates that as AI models become more integrated into daily workflows, their subjective qualities—their "vibe," reliability, and conversational style—are just as critical as their objective technical capabilities.[7][8]

In conclusion, OpenAI's introduction of "Auto," "Fast," and "Thinking" modes, alongside the reinstatement of user-selected models like GPT-4o, represents a significant recalibration. The company is actively seeking a new equilibrium between the elegant simplicity of an automated system and the empowering complexity of manual control. This evolution from a single, imposed experience to a more democratic, customizable interface acknowledges that the path forward for consumer AI involves not just building more powerful intelligence, but also fostering a more collaborative and responsive relationship with the humans who use it. This focus on user satisfaction and granular control may well define the next chapter of innovation and competition in the artificial intelligence landscape.