Meta Plans to Launch Standalone Meta AI App
Meta Platforms Inc. is reportedly planning to release a standalone mobile application for its Meta AI artificial intelligence assistant. According to a CNBC report, the app could debut as soon as the second quarter of this year. The Meta AI chatbot, similar to OpenAI’s ChatGPT, has been available since September 2023. Currently, it is integrated within Facebook, Instagram, WhatsApp, and Messenger and accessible through a dedicated website.
Speculation about a standalone app does align with earlier public sentiment. In January, a user on Threads suggested that Meta should develop a dedicated app to improve the chatbot’s usability, to which Meta Chief Executive Officer Mark Zuckerberg reacted with an emoji expressing agreement.
Potential Subscription Service
Alongside the mobile app, Meta is reportedly considering a paid subscription for Meta AI, although its launch date is uncertain. Details about the subscription’s additional features have yet to be released.
Offline Capabilities
A Threads post from last month also suggested including features such as “lightweight offline use” in the upcoming mobile version of Meta AI. Meta has already developed the foundational technology to support an offline mobile chatbot. Many AI apps rely on cloud-based large language models, requiring a network connection for functionality. To operate offline, chatbots must utilize a language model that is small enough to run directly on a user’s device.
Llama Models
In September, Meta released two lightweight models, Llama 3.2 1B and Llama 3.2 3B, that meet these requirements. These models are compatible with the systems-on-chip that power modern smartphones. When Llama 3.2 1B and Llama 3.2 3B were first introduced, Meta said they were optimized to run on Arm silicon, suggesting their possible integration into the Meta AI app. A few weeks later, Meta launched upgraded versions of these models that operated up to four times faster while using 41% less memory. The efficiency gains were achieved through a machine learning method called quantization, which involves compressing data before using it in a language model.
To minimize the reduction in output quality associated with compression, Meta developed a custom quantization algorithm called SpinQuant. Additionally, the company worked with Qualcomm and MediaTek to optimize quantized Llama 3 models for their mobile SoCs.
The Competition
It remains unconfirmed what features will be included in Meta’s expected paid subscription. Based on other chatbot apps, the subscription could offer higher usage limits or access to more sophisticated language models than the free version. One of these premium LLMs may be Meta’s upcoming Llama 4 algorithm, scheduled to launch later this year.
