OpenAI Responds to User Revolt with GPT-5 Control Overhaul
OpenAI has rolled out sweeping changes to GPT-5, delivering new speed modes, expanded message limits, and the return of GPT-4o after sustained user backlash. The update represents a clear pivotโฆ towards flexibility, with CEO Sam Altman acknowledging that recent restrictions had left power users feeling boxed in.
The changes come as ChatGPT commands over 900 million weekly active users globally, with Asia-Pacific markets driving significant adoption. The region accounts for substantial traffic shares, with India representing 9.06% of global usage and Japan contributing 4.11%, reflecting the model's growing importance across diverse professional landscapes.
Three-Tier Speed System Transforms User Experience
The introduction of Auto, Fast, and Thinking modes marks GPT-5's most significant usability upgrade. Auto serves as the balanced default, whilst Fast prioritises quick responses over depth. Thinking mode, however, is where the real innovation lies.
"The Thinking mode gives GPT-5 a formidable 196,000-token context windowโฆ, allowing it to process documents that previously needed splitting into chunks," explains Sarah Chen, AI researcher at Singapore's Institute for Infocomm Research.
For knowledge workers across Asia juggling multilingual contracts or extensive meeting transcripts, this expanded capacity removes a persistent workflow bottleneck. Weekly usage caps sit at 3,000 messages for Thinking mode, with a lightweight "mini" version providing overflow capacity.
The speed tiers address different professional needs effectively:
- Auto mode balances reasoning depth with response speed for general productivity tasks
- Fast mode delivers immediate answers for quick queries and brainstorming sessions
- Thinking mode handles complex analysis, technical documentation, and multi-step problem solving
- Mini overflow ensures continuous access when primary limits are reached
By The Numbers
- 831 million monthly users access ChatGPT globally as of March 2026
- 196,000-token context window in GPT-5 Thinking mode
- 3,000 weekly message limit for Thinking mode users
- 9.06% of global ChatGPT traffic originates from India
- 48.67% year-over-year growth makes ChatGPT the fastest-growing top-10 website
GPT-4o Returns After Developer Exodus
Perhaps the most telling concession involves GPT-4o's restoration to the default model picker. Its quiet removal earlier this year triggered immediate protests across developer forums, particularly in Asia where teams had built workflows around its specific performance characteristics.
The model's return signals OpenAI's recognition that choice matters more than streamlining. All paid subscribers now see GPT-4o by default, whilst a new "Show additional models" toggle unlocks access to o3, 4.1, and GPT-5 Thinking mini. Only GPT-4.5 remains restricted to Pro subscribers due to computational costs.
"We saw immediate productivity drops when GPT-4o disappeared from our development pipeline. Its balance of speed and reliability was irreplaceable for our multilingual content workflows," notes Kenji Nakamura, CTO at Tokyo-based fintech startup Zaiko.
This restoration acknowledges a fundamental truth: different models excel at different tasks, and forcing users into a single option reduces rather than enhances productivity. Teams working with complex project management scenarios particularly benefit from having multiple model options available.
Personality Customisation Addresses Cultural Nuances
OpenAI's promise of "warmer" personality defaults, coupled with eventual user customisation, addresses a uniquely Asian challenge. Professional communication styles vary dramatically across the region, from Singapore's analytical precision to Seoul's collaborative creativity.
The ability to tune AI personality represents more than convenience. A finance team in Hong Kong might prefer clipped, data-focused responses, whilst a Jakarta creative agency could benefit from more conversational, exploratory dialogue. This customisation capability positions GPT-5 as a genuinely adaptable tool rather than a one-size-fits-all solution.
Current personality improvements focus on reducing the divisive tone some users found off-putting in GPT-4o. The previous personality rollback demonstrated how sensitive users are to these changes, making gradual, user-controlled adjustments the smarter approach.
| Feature | Before Update | After Update |
|---|---|---|
| Speed Options | Single default mode | Auto, Fast, Thinking modes |
| Context Window | 128,000 tokensโฆ | 196,000 tokens (Thinking) |
| Model Choice | Limited selection | GPT-4o restored, additional toggle |
| Message Limits | Restrictive caps | 3,000 weekly + mini overflow |
| Personality | Fixed tone | Warmer defaults, customisation coming |
Market Pressure Forces Flexibility Focus
These changes arrive as competitors like Anthropic and Google DeepMind aggressively court enterprise customers across Asia with promises of better transparency and stronger guardrailsโฆ. Meanwhile, regulators in Japan and India scrutinise AI model governance with increasing intensity.
By emphasising user control over speed, depth, model selection, and personality, OpenAI signals it has absorbed criticism about being too restrictive. The updates position GPT-5 not as a monolithic assistant but as a flexible platform that adapts to diverse professional needs.
The timing is crucial. Teams exploring advanced AI thinking techniques need tools that match their workflows, not force workflow changes. Similarly, businesses developing AI-powered customer service solutions require predictable, customisable responses.
Regional adoption patterns suggest these flexibility improvements matter enormously. With India and Japan representing substantial user bases, cultural and professional diversity demands adaptable rather than prescriptive AI interactions.
What do the new speed modes actually do?
Auto balances depth and speed for general tasks, Fast prioritises quick responses, and Thinking provides extensive reasoning with a 196,000-token context window for complex analysis and document processing.
Why did OpenAI bring back GPT-4o?
User backlash was immediate and sustained after its removal. Developers and teams had built workflows around its specific performance characteristics, particularly for multilingual and technical content.
How many messages can I send in Thinking mode?
The limit is 3,000 messages per week for standard Thinking mode, with a lightweight "mini" version providing overflow capacity once you hit that ceiling.
When will personality customisation be available?
OpenAI hasn't provided specific timelines, but confirmed it's working on letting users dial in their preferred communication styles rather than accepting fixed defaults.
Do these changes apply to all subscription tiers?
Most features are available to paid subscribers, though GPT-4.5 remains exclusive to Pro subscribers due to higher computational costs. The additional models toggle works across paid tiers.
The broader question remains whether these changes address fundamental concerns about AI model governance and transparency, or simply paper over deeper issues with surface-level customisation. As Amazon's potential ยฃ8bn investment in OpenAI suggests, the stakes for getting this balance right extend far beyond user satisfaction.
Which aspect of these GPT-5 updates would transform your daily workflow most significantly: the expanded context window, restored model choice, or promised personality customisation? Drop your take in the comments below.
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