OpenAI’s GPT-5 Rollout Stirs Debate — and Huawei’s DeepSeek R2 Looms as a Challenger

When OpenAI released GPT-5 in the first week of October, the company billed the new product as the ultimate ChatGPT simplification. The concept sounded audacious: a single “size fits all” AI, governed by an intelligent router able to automatically determine whether a user requested lightening-speed answers or slow, deliberative thinking.

In theory, that meant no more hunting through a cluttered model picker menu — a feature CEO Sam Altman has openly admitted he dislikes. GPT-5’s router was supposed to handle it all, seamlessly blending speed, accuracy, and personality in a way that would make switching models obsolete.

But the dream started breaking in a few days.

Backlash and the Return of the Model Picker

Subscriber reaction was immediate — and not entirely kind. Subscribers complained about the inconsistent router of GPT-5, with some alleging the AI seemed slower or less powerful than predecessors. Paid users, in particular, resented the deletion of such personal favorites as o3, GPT-4o, and GPT-4.1, the latter two of which had cultivated personalities and styles appreciated by those who used them.

The bond was greater than technical; for others, emotional. AI aficionados have already started regarding models as old friends — so much so that San Francisco bid farewell to Anthropic’s retired Claude 3 Sonnet with a funeral this year.

Faced with mounting pressure, OpenAI reversed course. The model picker is back, but in a leaner form, with three selectable GPT-5 modes:

• Auto – allows the router to choose.
• Fast – prioritizes fast response times.
• Thinking – requires deeper, multi-step thinking.

For paid users, GPT-4o has been restored to the picker by default, while other legacy models can be enabled in settings. In a nod to transparency, Altman promised that future model retirements will come with advance notice.

Tuning the AI Personality

One unexpected lesson from the rollout: personality matters. Altman acknowledged GPT-5’s tone isn’t quite right yet, promising a “warmer” feel without the quirks some found irritating in GPT-4o.

More broadly, OpenAI is eyeing a future where personality isn’t one-size-fits-all. Instead, each user could fine-tune their AI’s style — from concise and formal to chatty and contrarian — in the same way people personalize playlists or phone themes today.

Technical Challenges Behind the Scenes

Behind the scenes, the routing isn’t straightforward. The system must recognize a person’s intention, reconcile it with the proper model, and return the response — all in milliseconds, without disrupting the conversational rhythm. Scalability isn’t merely about velocity. Certain users prioritize pithiness; others desire rich, even meandering excursions about a subject.

Thinking Mode of the GPT-5 pushes the boundaries even further, with 3,000 messages per week and a vast 196k-token context window — ideal for viewing large tech documents or managing large creative projects in one go.

Meanwhile in China: Introducing DeepSeek R2

Whilst OpenAI polishes up the GPT-5, there is a new contender about to enter the fray. Leaks point to DeepSeek R2, powered by Huawei's Ascend 910B AI processor, arriving sometime later this month.

Built on the success of the R1 model, R2 promises sharper reasoning, improved efficiency, and advanced logical processing — all while achieving 82% utilization of Huawei’s processors and a staggering 512 PetaFLOPS at FP16. If accurate, that puts it within striking distance of NVIDIA A100 cluster efficiency.

R2 is also expected to adopt an advanced Mixture of Experts (MoE) architecture, allowing it to dynamically route tasks to specialized “expert” networks for heavy-duty reasoning without wasting compute power.

Notably, DeepSeek appears committed to cost-effectiveness, open-source principles, and operational efficiency, which could make it attractive to developers wary of expensive, closed ecosystems.

The Competitive Landscape Is Heating Up

If DeepSeek R2 lives up to the spec leaks, the world of AI may see renewed competition. A rocky launch for GPT-5 provides some opportunities for competitors with a narrow window of getting attention — not to mention both raw power alongside the personal touch users appear to want from their AI.

OpenAI, for its part, isn’t standing still. By restoring legacy models, offering more control over speed and reasoning depth, and exploring personality customization, it’s signaling a willingness to adapt quickly — a vital trait in a market evolving at breakneck speed.

Conclusion: The Age of AI Personalization

The release of GPT-5 demonstrates the AI development isn’t so much about making the model more intelligent; it’s about making the model yours. Users desire speed, accuracy, power of reasoning — but also personality, predictability, and the familiarity of tried-and-tested styles. With Huawei's DeepSeek R2 about to take on GPT-5 on performance, the next front could well be emotional alignment — not merely technological ability. Ultimately, the most effective AI may not be the one with the sharpest mind, but the one with the right feel for the user.