Apple has released the latest update for its artificial intelligence models, which primarily supports the Apple Intelligence feature across systems like iOS and macOS. According to Apple's official data, the newly launched model performs comparably to similar products from Google and Alibaba in terms of performance. However, when compared to OpenAI’s GPT-4o, which was released a year ago, Apple's server-side model shows significantly weaker performance.
In the update, Apple emphasized the capabilities of its "Apple Device-Side Model." This model can run on devices like iPhones without an internet connection, and its text generation quality, tested by human evaluators, showed no significant difference compared to similar-scale models from Google and Alibaba. However, the results for the "Apple Server Model" were less favorable, failing to outperform OpenAI's earlier offerings in multiple tests.
In image analysis capabilities, Apple also struggled to stand out. According to Apple’s own data, human evaluators tended to prefer Meta's Llama4Scout model over Apple's server-side product, despite Llama4Scout not being particularly outstanding in several tests. This result came as a surprise and highlighted the competitive pressure Apple faces in this field.
The updated "Apple On-Device" model contains approximately 3 billion parameters, primarily for text generation, summarization, and text analysis functions. The number of parameters generally correlates with the model's problem-solving ability; more parameters usually mean better performance. Apple stated that the new model has improved tool usage and efficiency compared to previous versions, supporting around 15 languages. This is thanks to its expanded training dataset, which includes diverse data such as images, PDF files, documents, and various types of charts.
Despite Apple's efforts in the AI field, the market remains highly competitive. Many users expressed disappointment with Apple's progress on personalized Siri upgrades, and delays in these updates have caused dissatisfaction among some consumers. This even led to lawsuits filed by certain users accusing the company of failing to deliver on its promised AI features.
Key Points:
🌟 Apple's device-side model performs comparably to Google and Alibaba, but its server-side model falls short of OpenAI’s GPT-4o.
🖼️ In image analysis, Apple's model underperforms Meta's Llama4Scout.
🔧 The new "Apple On-Device" model has about 3 billion parameters and supports various text processing functions.
