Apple’s upgraded AI models underwhelm on performance

Apple Intelligence features within Apple, iOS, MacOS and more announced the updates to AI models that strengthen their group. However, according to the company’s own benchmarks, models, including Openai, the upside down models of old models from rival technological companies.

Apple said in a blog post Monday, human tests, iPhone, including the latest “Apple on-device” model, not better than “relatively” but not better than Google and Alibaba models. Meanwhile, the same testers evaluated the more new model called Apple’s “Apple Server” and are designed to operate Openai behind the elderly GPT-4O.

Apple Models assessing the ability to analyze images, apple, Apple’s Llama 4 scout model preferred to the Apple server. This is a little striking. In a number of tests, 4 Call Scout Google operates worse than leading models from AI laboratories such as Anthropic and Openai.

The assessment results add validity to reports offering Apple’s AI research section. Apple’s AI opportunities in recent years fradmedAnd the promised SECORATION update was postponed for an indefinite period. Some customers The apple was suedTo accuse the AI ​​features for the products that the marketing fires still did not deliver.

In addition to creating text, the Apple, which is about 3 billion parameter of 3 billion, manages the features such as generalization and text analysis on the device. (Settings matches a model problem solving skills and models with more parameters generally perform better than more parameters.)

Apple boasts the use and efficiency of an improved tool compared to both apples and the predecessors of the Apple server’s predecessors and understand about 15 languages. Thank you partly to the extended training database that includes information, pdf, documents, manuscripts, infographics, tables and graphs.