Apple announced updates to its AI model

Apple announced yesterday at the WWDC 2025 upcoming updates on its AI model to feature across the intelligence suite. The model underperformed on its benchmark compared to old models from rival firms such as OpenAI.However, the tech giant argued that the On-Device and Apple Server have been improved compared to their predecessors and can understand more than 15 languages.

According to human testers, the newest on-device model that runs locally on iPhones and other devices performed similarly to old models from Google and Alibaba regarding text generation. Human testers rated Apple Server, designed to run in the company’s data centers, behind OpenAI’s year-old GPT-4o. In a separate evaluation, they favored Meta’s Llama 4 over Apple Server in the ability to analyze images. The ratings surprised Apple since Llama 4 Scout performed poorly compared to AI labs like Google, Anthropic, and OpenAI.

Developers feel skeptical about Apple’s promise of innovation

Developers feel negatively towards these years’ WWDC releases, including Apple’s small On-Device AI model. Atul Kakkar, chief product officer of AI-powered personal finance app Origi, revealed that as a developer, he did not feel the intensity of excitement around what Tim Cook’s company released.  He believes the annual WWDC event was once tech’s buzziest affair, but now it’s surrounded by skepticism that the iPhone maker can keep pace with AI giants.

Apple doesn't report benchmarks for their AIs, reporting on an ill-documented head-to-head evaluation

But even by their standards, Apple's latest on device models are mostly worse than the open Gemma 3-4B from Google or Qwen 3-4B

And their server LLM is similar to Llama 4 Scout pic.twitter.com/hwpOG5Lgyp

— Ethan Mollick (@emollick) June 10, 2025

The tech giant is under pressure to keep up by continually updating its AI tools for developers and building a thrilling consumer experience within the Apple Intelligence suite. Grant McDonald, chief executive and co-founder of the AI-powered parenting advice app Bobo, said that Tim Cook’s company is doing interesting things but is not enough to break through the noise.

McDonald believes that letting developers build AI-powered applications with its On-Device small AI models could be the value the company hopes to deliver. Small AI models are trained on small datasets and use less computing power, meaning they can be run on an iPhone rather than large servers in the cloud. He said running models on a device is practically cheaper than cloud-based ones, such as GPT and Gemini. 

Developers have doubted that On-Device models can perform well compared to cloud-based ones. One developer said they won’t be able to fine-tune them on a proprietary dataset. McDonald said the on-device models will be good enough for some apps but not for Bobo. The parenting app customized its AI model with proprietary data from sleep consultants, lactation consultants, speech pathologists, and other parenting experts. 

Apple’s AI research division struggles to catch up with competitors

The tech giant promised a Siri upgrade to feature more personalized content, but this upgrade has been delayed. Some customers have sued the company for marketing products that have not yet been delivered. Researchers said that while AI models perform very well in math and science, complex problems provide just the illusion of thinking. 

The tech giant delayed developing large language models and implementing AI in its devices to stay out of the conversation. At WWDC, the company unveiled Apple Intelligence for AI features, which the community considered underwhelming. 

Researchers have figured out that LRMs start to fail after a certain point. It shows that all reasoning models exhibit a similar pattern in which accuracy progressively declines as problem complexity increases until zero accuracy. According to Researchers, Reasoning models initially apply more thinking tokens, but as complexity increases, they give up, and counterintuitively, their reasoning effort begins to reduce despite increasing problem difficulty. 

According to Dipanjan Chatterjee, a Forrester Research Analyst, the silence surrounding Siri was deafening. He revealed that no amount of text correction or cute emojis can fill the awaited interactive AI experience that he knows Siri would be capable of. He believes that the end for Siri is near if the firm does not act fast. The company did not provide a precise timetable for Siri’s upgrade. Still, it indicated that it won’t happen until next year at the earliest, according to Craig Federighi, a top software executive in the company.

Your crypto news deserves attention - KEY Difference Wire puts you on 250+ top sites