OpenAI has rolled out a fresh wave of changes to GPT-5, giving users more say over how they use the model. The update introduces new speed modes, restores GPT-4o to the model picker, and lifts usage limits for those who lean heavily on the “Thinking” mode. Sam Altman, OpenAI’s CEO, framed the tweaks as a bid to improve flexibility without sacrificing performance. The announcement followed a flurry of criticism from long-time users who felt recent changes left them with fewer choices.
New speed modes let users switch between Auto, Fast and Thinking, balancing quick replies with more intensive reasoning.,Message caps expanded, with 3,000 weekly messages for GPT-5 Thinking and a new “mini” version for overflow.,Model picker revamped, with GPT-4o restored for paid users after backlash, plus options to toggle additional models.
Why the speed modes matter
The introduction of Auto, Fast and Thinking represents a subtle but important shift. Auto, the default, balances depth with pace, while Fast is designed for snappier exchanges. Thinking, meanwhile, leans into reasoning-heavy requests such as analysing long reports or drafting technical arguments.
OpenAI has given the mode a formidable 196,000-token context window, which allows GPT-5 to keep track of far longer conversations or process documents that once would have needed to be broken into parts. For knowledge workers in Asia juggling multilingual contracts, financial data, or extended meeting notes, this kind of capacity will be a noticeable upgrade.
Weekly usage is capped at 3,000 messages, though a lightweight GPT-5 Thinking mini provides overflow once that limit is hit. OpenAI hinted that these ceilings could change as it learns from usage patterns.
The quiet return of GPT-4o
Perhaps the most significant concession to user feedback is the return of GPT-4o, which had been quietly retired from the default model picker earlier this year. Its removal sparked immediate frustration across developer forums and productivity groups in Asia, where teams had tuned workflows around its distinct balance of speed and reliability.
The reinstatement means all paid accounts will now see GPT-4o by default. For those wanting to explore further, a new “Show additional models” toggle in ChatGPT’s web settings unlocks a menu of options including o3, 4.1, and GPT-5 Thinking mini. GPT-4.5, however, remains the preserve of Pro subscribers due to its high GPU costs.
Personality updates and customisation ahead
Altman also confirmed that GPT-5’s personality will evolve to feel “warmer”, while avoiding the divisive tone that some found off-putting in GPT-4o. The longer-term ambition is to let users dial in their own preferences — a clear nod to the reality that one person’s “engaging” assistant can feel flippant or distracting to another.
This ability to tune personality could be particularly valuable across Asia’s diverse professional landscape. A finance team in Singapore might prefer a clipped, analytical voice, while a creative agency in Seoul could want something looser and more conversational. The prospect of per-user customisation brings OpenAI closer to that level of adaptability.
The bigger picture
The changes arrive at a delicate moment for OpenAI. Rivals such as Anthropic and Google DeepMind are pitching their own large language models to enterprise customers in the region, often with promises of stronger guardrails or better transparency. Meanwhile, regulators in countries like Japan and India are scrutinising how AI models handle data and output. By giving users more control — over speed, depth, model choice and tone — OpenAI is signalling that it has heard the frustrations. More importantly, it is trying to position GPT-5 not as a one-size-fits-all assistant, but as a flexible platform that can flex to the many ways professionals across Asia are beginning to use it. For more insights into how these models are being adopted, check out our article on How People Really Use AI in 2025.
Whether these changes will be enough to steady the ship remains to be seen. But one thing is clear: GPT-5 is no longer just about raw intelligence. It is about choice, variety and responsiveness — the very things that may prove decisive in an increasingly crowded market. For a broader perspective on the AI landscape, you might find our discussion on AI's Secret Revolution: Trends You Can't Miss insightful. The debate around what constitutes "responsible innovation" in AI is also gaining traction, particularly in places like Taiwan, as explored in Taiwan’s AI Law Is Quietly Redefining What “Responsible Innovation” Means. For further reading on the societal implications of such models, a report by the National Academies of Sciences, Engineering, and Medicine provides valuable context on responsible development and use of AI^ https://nap.nationalacademies.org/catalog/26815/responsible-development-and-use-of-powerful-ai-systems.
Do you prefer speed, depth, or personality in your AI assistant — and which of these new GPT-5 features would you actually use day to day?
Latest Comments (2)
The 196,000-token context window for Thinking mode sounds good on paper, but how does that translate for real-world multilingual contracts here? We often deal with code-switching and complex legal jargon across bahasa and english. Will it handle that nuance effectively, or just give us a long but ultimately generic output?
The reintroduction of GPT-4o following user feedback, while seemingly minor, does underscore the industry's responsiveness. Such adjustments are valuable precedents, especially as the UK AI Safety Institute begins its scrutiny of frontier models.
Leave a Comment