OpenAI Expands GPT-5 with Four New Simplified Modes

OpenAI introduced its recently released GPT-5 as “a more efficient model that responds optimally to user questions.” Unlike previous models, users no longer had to choose between a quick response or deep thinking; the AI made the decision and responded accordingly. However, after a series of user complaints about GPT-5, OpenAI came to the realization that “simple is not always better.” Four new modes have been added to GPT-5. OpenAI CEO Sam Altman announced the introduction of these new modes in a post on “X.” The existing “Auto” mode remains, intended for users who want the AI to decide how to respond. Meanwhile, new modes have been added: “Fast” mode (quick response), “Thinking” mode (deep thinking), and “Thinking mini” mode (short thinking). Altman stated, “While most users will likely use Auto, the added controls will be useful for some.” GPT-5, released just a week ago, was designed by OpenAI as a “unified system without the need for mode selection.” Users simply enter prompts, and ChatGPT automatically determines the response style.

When introducing GPT-5, OpenAI explained, “GPT-5 is a unified system with a smart and efficient model for answering most questions, a deep inference model (GPT-5 Thinking) for challenging problems, and a real-time router that instantly determines which model to use based on the type and complexity of the conversation, tool needs, and the user’s explicit intent.” Furthermore, this router continuously learns and improves based on actual signals, such as when users switch models, answer preferences, and accuracy measurements. However, this “elimination of complexity” approach ultimately drew significant criticism. Following the release of GPT-5, OpenAI received harsh criticism. Much of this criticism lamented the sudden disappearance of GPT-4, with many users feeling that GPT-5 was less powerful than its predecessor. Interestingly, these criticisms were emotional, with some expressing sadness as if they had lost a “close friend” they trusted and relied on. In response to these reactions, OpenAI reinstated the older models, including GPT-4o, GPT-4.1, o4-mini, and o3.

However, these are only available to paid Plus and Pro users, while free users will continue to use GPT-5. Similarly, the four new modes of GPT-5 are also available only to paid users. As a result, ChatGPT has reverted from a “simple, unified process” to a “complex system with multiple options.” Users are now forced to decide whether to use GPT-5, GPT-4o, o4-mini, or o3, and, if they choose GPT-5, whether to leave it in Auto mode or select a specific mode.

こちらもお読みください： ChatSense “o3” Adds Web Search with “Search Agent”

In his X post, Altman said, “4o reverts to the model picker by default for all paid users. If we decide to discontinue it in the future, we will provide ample notice. Paid users can add models such as o3, 4.1, and GPT-5 Thinking mini by using the ‘Show additional models’ toggle in ChatGPT’s web settings. Note that 4.5 is only available to Pro users and uses significantly more GPUs.” Given the recent rapid changes, オープンAI appears to be exploring how to handle GPT-5: should it be simple, complex, or somewhere in between? Altman’s post ended with a comment that hinted at ChatGPT’s future direction. “We’re currently updating GPT-5’s personality to be warmer than its current personality and less intrusive (for many users) than GPT-4o. However, one lesson we’ve learned over the past few days is that we need to move towards a world where model personalities can be more customized for each user.” This article was originally published by Ziff Davis overseas and edited for Japan by Asahi Interactive.

ソース ヤフー