Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, information, and safety leaders. Subscribe Now
The launch of OpenAI’s lengthy anticipated new mannequin, GPT-5, is off to a rocky begin to say the least.
Even forgiving errors in charts and voice demoes throughout yesterday’s livestreamed presentation of the brand new mannequin (truly 4 separate fashions, and a ‘Pondering’ mode that may be engaged for 3 of them), a variety of consumer studies have emerged since GPT-5’s launch displaying it erring badly when fixing comparatively easy issues that previous OpenAI fashions — and rivals from competing AI labs — reply accurately.
For instance, information scientist Colin Fraser posted screenshots displaying GPT-5 getting a math proof fallacious (whether or not 8.888 repeating is the same as 9 — it’s after all, not).
It additionally failed on a easy algebra arithmetic drawback that elementary schoolers might in all probability nail, 5.9 = x + 5.11.
AI Scaling Hits Its Limits
Energy caps, rising token prices, and inference delays are reshaping enterprise AI. Be part of our unique salon to find how prime groups are:
- Turning power right into a strategic benefit
- Architecting environment friendly inference for actual throughput beneficial properties
- Unlocking aggressive ROI with sustainable AI methods
Safe your spot to remain forward: https://bit.ly/4mwGngO
Utilizing GPT-5 to evaluate OpenAI’s personal misguided presentation charts additionally didn’t yield useful or right responses.
It additionally failed on this trickier math phrase drawback under (which, to be honest, stumped this human at first…although Elon Musk’s Groq 4 AI answered it accurately. For a touch, consider the truth that flagstones on this case can’t be divided into smaller parts. They have to stay in tact as 80 separate models, so no halves or quarters).
Not pretty much as good at coding as benchmarks point out
Regardless that OpenAI’s inner benchmarks and a few third-party exterior ones have proven GPT-5 to outperform all different fashions at coding, it seems that in actual world utilization, Anthropic’s lately up to date Claude Opus 4.1 appears to do a greater job at “one-shotting” sure duties, that’s, finishing the consumer’s desired software or software program construct to their specs. See an instance under from developer Justin Solar posted to X :
Sadly, OpenAI is slowly deprecating these older fashions — together with the previous default GPT-4o and the highly effective reasoning mannequin o3 — for customers of ChatGPT, although they’ll proceed to be obtainable within the software programming interface (API) for builders for the foreseeable future.
As well as, a report from safety agency SPLX discovered that OpenAI’s inner security layer left main gaps in areas like enterprise alignment and vulnerability to immediate injection and obfuscated logic assaults.
Whereas anecdotal, the checking the temperature on how the mannequin is faring with early AI adopters appears to point a cold reception.
AI influencer and former Googler Bilawal Sidhu posted a ballot on X asking for a “vibe examine” from his followers and the broader userbase, and up to now, with 172 votes in, the overwhelming response is “Kinda mid.”
And because the pseudonymous AI Leaks and Information account wrote, “The overwhelming consensus on GPT-5 from each X and the Reddit AMA are overwhelmingly adverse.”
Tibor Blaho, lead engineer at AIPRM and a preferred AI leaks and information poster on X, summarized the numerous issues with the ChatGPT-5 rollout in a superb publish, highlighting that one of many new marquee options — an automated “router” in ChatGPT that chooses a considering or non-thinking mode for the underlying GPT-5 mannequin relying on the issue of the question — has turn out to be one of many chief complaints, given the mannequin appeared to default to non-thinking mode for a lot of customers.
Competitors ready within the wings
Thus, the sentiment towards ChatGPT-5 is much from universally optimistic, highlighting a serious drawback for OpenAI because it faces growing competitors from main U.S. rivals like Google and Anthropic, and a rising checklist of free, open supply and highly effective Chinese language LLMs providing options that many U.S. fashions lack.
Take the Alibaba Qwen Crew of AI researchers, who simply at present up to date their extremely performant Qwen 3 mannequin to have 1 million token context — giving customers the flexibility to change practically 4x as a lot info with the mannequin in a single again/forth interplay as GPT-5 presents.
Given OpenAI’s different huge launch this week — that of latest open supply gpt-oss fashions — additionally obtained a combined reception from early customers, issues usually are not wanting up for the primary devoted AI firm by customers proper now (700 million weekly energetic customers of ChatGPT as of this month).
Certainly, that is additionally exemplified by customers of the betting market Polymarket overwhelmingly deciding following the discharge of GPT-5 that Google would seemingly have the perfect AI mannequin by the top of this month, August 2025.
Different energy customers like Otherside AI co-founder and CEO Matt Schumer, who obtained early entry to GPT-5 and blogged about it favorably in a overview right here, opined that views would shift as extra individuals discovered the perfect methods to make use of the brand new mannequin and adjusted their integration approaches:
Whereas it’s nonetheless early days for GPT-5 — and the sentiment might change dramatically as extra customers get their fingers on it and take a look at it for various duties — the early indications usually are not wanting like this can be a “house run” launch for OpenAI in the identical approach that prior releases akin to GPT-4, and even the newer 4o and o3, had been. And that’s a regarding indicator for an organization that simply raised one more funding spherical, but stays unprofitable on account of its excessive prices of analysis and growth.