Sam Altman Faces Backlash Over GPT-5 Performance as OpenAI Promises Fixes

OpenAI’s latest flagship model, GPT-5, is being met with mixed reviews. While blisteringly fast in coding tasks, early users say it underdelivers in creativity and reasoning — two areas CEO Sam Altman promised would shine. Now, an update is quietly rolling out to address those gripes.

Early Hype Meets Frustrated Users

When GPT-5 began its phased rollout, anticipation was sky-high. Altman had described it as a leap forward, hinting at better reasoning, longer answers, and more natural creativity. Instead, many testers say it’s been, well… a little flat.

In performance checks by BleepingComputer, GPT-5 blew past earlier OpenAI models in coding, often completing tasks at almost twice the speed of its predecessor and even beating the high-efficiency o3 model. But outside the programming sandbox, results dipped sharply.

One common complaint? It keeps answering in short bursts, even when a detailed, multi-paragraph breakdown is requested. For creative writing, it can feel mechanical — almost like it’s holding back.

The “Broken Autoswitcher” Bug

Altman insists it’s not intentional. In a post on X, he explained that part of GPT-5’s “autoswitcher” — the behind-the-scenes system that activates its reasoning capabilities — malfunctioned for hours.

“Yesterday, the autoswitcher broke and was out of commission for a chunk of the day, and the result was GPT-5 seemed way dumber,” Altman wrote.

He’s framing it as a technical hiccup, not a strategic cost-cutting measure. Still, some users are unconvinced, speculating that output length may be capped to keep server bills in check.

Rate Limits, Legacy Models, and a Slow Rollout

Altman has promised that Plus subscribers will soon see double the GPT-5 rate limits — but warns it’ll take several days to implement. In the meantime, users will be given a choice to stick with GPT-4o if they prefer.

A brief breakdown of what’s changing:

• Rate limits: Plus users will get double the allowance.

• Legacy models: GPT-4o will return for those who want it.

• Interface tweaks: Clearer indicators showing which model is responding.

Right now, GPT-5 is still being distributed across both free and paid tiers, meaning your account might not have it yet. Pro users on the $200 plan have exclusive access to older models, but Altman’s remarks suggest this may shift as demand is evaluated.

Creativity Still a Sticking Point

Speed alone doesn’t sell an AI model to everyone. Writers, marketers, and researchers say GPT-5 sometimes struggles to “think” in more imaginative or nuanced ways. Even when prompted with open-ended tasks, it can default to formulaic phrasing.

Some have tested it side-by-side with GPT-4, reporting that the older model occasionally outperforms its successor in generating engaging narratives or persuasive arguments. That’s an awkward comparison for a system billed as a generational leap.

In one case, a user fed both models the same creative brief. GPT-4 returned a vivid, layered story with character arcs. GPT-5 offered a short, functional outline — technically correct, but emotionally flat.

The Reasoning Toggle

To counter that, OpenAI is experimenting with a “reasoning mode” toggle. This will allow users to force GPT-5 into its deeper thinking process — the same one that’s supposed to kick in automatically but hasn’t been doing so reliably.

Altman says the company is also redesigning the interface so users can clearly see when GPT-5 is engaging that reasoning stack.

A small but significant change is planned: making it easier to manually trigger “thinking” responses without relying on hidden system decisions. If the update works as intended, it could address one of the most frustrating aspects for power users.

Why It Matters for OpenAI’s Strategy

For OpenAI, GPT-5’s launch is more than just another model release — it’s a flagship moment in a competitive AI market where Google, Anthropic, and smaller startups are all trying to out-innovate each other. A slow start risks ceding momentum.

The performance gap between coding excellence and creative underperformance raises strategic questions. Does OpenAI double down on what GPT-5 is already great at, or push through on making it a balanced all-rounder?

There’s also the matter of user trust. Public beta rollouts can be bumpy, but overpromising and underdelivering can erode goodwill, especially among paying customers.

The Road Ahead

OpenAI’s engineering teams are pushing out incremental fixes over the coming days. Altman has said GPT-5 will be “smarter starting later today,” though he hasn’t given a specific timetable for full stability.

For now, GPT-5 feels like a high-performance sports car that sometimes forgets how to shift gears. When it’s in the right mode, it’s blisteringly good. When it’s not, it can feel oddly sluggish in the areas it was supposed to shine.

If the new toggle, rate limit boost, and UI improvements work as intended, GPT-5 could start living up to the hype. But until then, OpenAI has some convincing to do — and plenty of eyes watching closely.