PrismML overcomes the bottleneck of edge AI: successfully integrates Ali Tongyi Qianwen into Apple Inc. (AAPL.US) iPhone 17 Pro, running all parameters comparable to the cloud.

According to informed sources, the artificial intelligence startup PrismML has compressed the open-source large language model Qwen 3.6 of Alibaba Group Holding Limited Sponsored ADR (BABA.US) to make it run on the iPhone 17 Pro of Apple Inc. (AAPL.US). The sources indicated that Apple Inc. has held a meeting with PrismML to discuss how to use its technology. The model has 27 billion (27B) parameters, which can help determine the complexity of the data the model can handle. In contrast, most models running on phones have only a few billion parameters activated at a time. The largest artificial intelligence model can have trillions of parameters, which is still too large to run on mobile devices. However, according to the startup, PrismML, the model that successfully runs on the iPhone can perform complex tasks such as dialogue, reasoning, fully autonomous intelligent agents, and software code writing. The report also mentioned that the open-source model will be available for download next Tuesday. PrismML uses a mathematical technique to shrink the volume of the Qwen 3.6 model to a small fraction of its original size. Usually, shrinking a model can lead to a decrease in performance, but the company claims that its method of miniaturizing artificial intelligence models does not compromise its performance. It is reported that PrismML has compressed the volume of Qwen 3.6 from approximately 54 gigabytes (GB) to less than 4 GB. This milestone event reflects a broader trend of pushing artificial intelligence to run on devices rather than expensive high-power servers in data centers. Companies like Microsoft Corporation (MSFT.US), Amazon.com, Inc. (AMZN.US), Meta Platforms (META.US) are investing billions of dollars in building data centers to meet their anticipated demand for artificial intelligence. However, Apple Inc. has largely remained on the sidelines of this data center race while also being a staunch advocate of ensuring that many of the artificial intelligence features on the iPhone run on the device side rather than in the cloud. PrismML's approach may appeal to Apple Inc. At the company's Worldwide Developers Conference (WWDC) in June this year, Apple Inc. announced a comprehensive overhaul of Siri based on the Alphabet Inc. Class C (GOOGL.US) Gemini model. The most advanced parts of Siri are still too large in volume, requiring Apple Inc. to use chips from NVIDIA Corporation (NVDA.US) running in the Alphabet Inc. Class C cloud. As part of the new Siri release, Apple Inc. stated that some new AI features on the iPhone will run on the device side. Apple Inc.'s new on-device model has 20 billion (20B) parameters but uses a sparse architecture, where only 1 billion to 4 billion (1B to 4B) parameters are activated at a time. Sources add that in PrismML's on-device model, all 27 billion parameters are activated simultaneously. As the first venture capital investor in OpenAI, Khosla Ventures invested $16.25 million in seed funding for PrismML earlier this year. PrismML is a spin-off from the California Institute of Technology, with its CEO, Babak Hassibi. Hassibi is a professor of electrical engineering at the university and conducted the mathematical research used by the startup's technology. The report notes that the California Institute of Technology holds the patent behind the technology but has exclusively licensed it to PrismML.