2025 was supposed to be the year of AI breakthroughs. Instead, several highly anticipated launches left users more frustrated than impressed. While many of these products were not outright failures or flops after launch and fixes, the initial excitement could not be delivered, leaving users underwhelmed.
From bots that fell short of their promises to apps that simply weren’t functional, here’s a rundown of the year’s biggest AI letdowns based on real user feedback.
OpenAI’s GPT-5
OpenAI’s GPT-5 launch in August quickly became a spectacle of disappointment. CEO Sam Altman compared it to ‘Death Star’, a superweapon from the Star Wars franchise. Instead, many users felt misled.
Forcibly retiring GPT-4o, GPT-4.5 and GPT-3 overnight sparked outrage, with over 3,000 users petitioning to bring them back. GPT-5 was slower, gave shorter and less detailed answers, lost personality and warmth, and felt more restricted.
Reddit users slammed it as a “downgrade branded as the new hotness”. Even Altman later admitted OpenAI “screwed up” the rollout. Now, OpenAI has taken several steps to fix GPT-5 and is also planning to allow users to use older models for specific tasks.
Meta Llama 4
Meta surprised everyone with Llama 4 in April, but developers quickly found that it underperformed expectations. Contrary to expectations, it was not truly open source. Instead, it arrived as an open weights model with several usage restrictions.
Llama 4 struggled on multiple benchmarks, scoring only 16% on Aider’s polyglot coding tests, barely matching smaller models. Its long-form writing performed even worse than its smaller peers, while the initial model sounded juvenile and inconsistent.
Users accused Meta of benchmark manipulation and public defences by Meta executives couldn’t repair the damage.
Sarvam-M
Launched in May as India’s first model built under the IndiaAI Mission, Sarvam-M struggled to make an impact in its initial launch.
Despite claims of strong performance in Indian languages and math, it only got 334 downloads in two days. Users said it was “not quite at ChatGPT’s level”—adequate for basics but shallow overall. Investors called the response “embarrassing”, while social media backlash questioned what the model had actually achieved.
Cut to today, Sarvam-M has crossed lakhs of downloads on Hugging Face and thousands on AIKosh. The company is also planning to release India’s first foundational model from scratch in early 2026.
Perplexity Pro with Airtel
Airtel promised 12 months of Perplexity Pro free to millions of customers, but the reality didn’t quite live up to expectations. While Perplexity has indeed been freely accessible to several researchers and users in India, those under the Airtel plan felt that the models used were not the same as the paid Pro version.
Airtel’s ‘Pro’ version ended up performing worse than Perplexity’s free version. Responses were slower, citations were missing, visuals were absent and insights were shallow. Users felt misled because there was no disclaimer that it was a limited version. Perplexity dismissed complaints, but the experience spoke for itself.
Apple Intelligence
Apple’s AI initiatives throughout the year failed to impress or drive adoption. Though Apple announced it last year, people were not convinced by the announcement.
Promised features, like Siri searching through emails and notes, didn’t materialise. Summarisation tools were riddled with errors, including one notorious case where they generated wildly inaccurate news summaries, forcing the company to roll back the feature.
Users called Siri “underwhelming for years” and Apple’s stock apps “mediocre”, questioning whether any real AI upgrade had arrived at all.
Taco Bell AI Drive-Through
Taco Bell’s AI voice ordering, rolled out across 500 locations, quickly became a source of frustration rather than convenience.
The AI system accepted orders for 18,000 cups of water, got stuck in loops and added unwanted extras to meals. Customers complained of duplicate charges. Taco Bell eventually acknowledged the failures, slowing adoption and forcing human intervention.
Ola Krutrim’s Kruti
Krutrim Kruti launched as India’s answer to agentic AI, but the launch faltered. Bhavish Aggarwal’s Krutrim AI startup has been on the radar for most of its AI announcements over the last two years, yet users consistently felt the model could not live up to the hype created during every launch.
Queries were slow or incorrect, often repeating scripted responses. Downloads were low, layoffs occurred, executives resigned and fundraising goals had to be scaled back. Evidence suggested that Krutrim Kruti appeared to be little more than Llama 3 or Mistral models wrapped with heavy prompting, far from the revolutionary product it promised to be.
Meta Vibes
Meta’s TikTok-style AI video feed, Vibes, launched in September, flopped almost immediately.
Users dismissed it as ‘AI slop’, filled with meaningless, low-quality content. The platform contradicted Meta’s own claims about authentic storytelling, instead flooding users’ feeds with unappealing AI-generated videos. This drew widespread ridicule and criticism.
Sora 2 App
Similar to Vibes, OpenAI’s Sora 2 proved that technical AI improvements couldn’t rescue a weak social experience.
App ratings fell to 2.9 stars. Strict content restrictions, limited generations for paid users, cut-off videos and buggy animations made the app frustrating. Users described it as “boring”, “unpredictable, expensive, overhyped and underwhelming”, often finding videos unusable for posting.
Runway Gen-4
After repeated delays, Runway Gen-4 arrived with promises of big improvements, but premium users were left disappointed.
Queue times exceeded 10 minutes, animated characters behaved erratically and videos often froze awkwardly before jolting into action. Many users argued that Gen-3 Alpha Turbo was still superior, prompting cancellations of subscriptions in favour of rivals like Kling and Sora.
2025 made one thing clear — hype doesn’t guarantee quality. These AI launches promised the moon, but for many users, they barely left the ground.
The post Top 2025 AI Launches That Were as Underwhelming as Overhyped appeared first on Analytics India Magazine.
