The UAE’s Falcon 3 is challenging open source leaders amid demand for smaller AI models

The UAE government supports it Technology Innovation Institute (TII) has announced the launch of Falcon 3, a family of small open source language models (SLMs) designed to run efficiently on lightweight, single GPU-based hardware.

Falcon 3 has four models – 1B, 3B, 7B, and 10B – with basic and training capabilities, and promises to democratize access to advanced AI capabilities for developers, researchers, and businesses. According to Hugging Face’s leadership team, these models are outperforming or very similar to their well-resourced counterparts in their class, including Meta’s Llama and team leader Qwen-2.5.

Development comes in time importance of SLMswith fewer parts and a simpler design than LLMs, it is growing rapidly due to its ability, affordability, and ability to be used on unlimited devices. It is suitable for use in various industries, such as customer service, healthcare, mobile applications and IoT, where conventional LLMs can be too expensive to manage. According to Appreciate the Reportsthis type of market is expected to grow, with a CAGR of about 18% over the next five years.

What does the Falcon 3 bring to the table?

Trained on 14 trillion tokens – more than double its predecessor Falcon 2 – the Falcon 3 family uses automatic query-aware architectures to partition parameters and reduce the memory usage of the virtual value (KV) during prediction. This helps to work quickly and efficiently when performing various voice-based tasks.

At the core, the models support four primary languages ​​- English, French, Spanish, and Portuguese – and have a 32K screen, allowing them to process long inputs, such as text.

“Falcon 3 is versatile, designed for both routine and special tasks, which offers great flexibility to users. Its basic model is perfect for production use, while the training model excels in discussions as a customer or agent,” TII says. website.

According to board board on Hugging Face, while all four models of Falcon 3 work well, versions 10B and 7B are the stars of the show, achieving modern results in thinking, understanding language, following instructions, codes and mathematics.

Among the models under the 13B segment size class, Falcon 3’s 10B and 7B models outperform its competitors, including Gemma 2-9B by GoogleMeta’s Llama 3.1-8B, Mistral-7Band Yi 1.5-9B. It outperforms the group leader Alibaba Qwen 2.5-7B in many benchmarks – such as MUSR, MATH, GPQA, and IFEval – except for the MMLU, which is a test to assess how well linguists understand and process human language.

Industrial deployment

It’s the Falcon 3 models that are available now Hugging FaceTII aims to be user-friendly, enabling low-cost AI deployment without computational complexity. With their specialized, domain-focused capabilities and fast turnaround times, the brands can leverage a variety of applications at the edge and in privacy-sensitive environments, including customer service chatbots, personalization systems, data analytics, fraud detection, health assessment, service optimization and education.

The agency also plans to expand the Falcon family beyond introducing multimodal models. These models are expected to be launched sometime in January 2025.

In particular, all models have been released under the TII Falcon License 2.0, an official Apache 2.0 license with an official usage policy that promotes AI development and deployment. To help users get started, TII has also launched the Falcon Playground, a testing environment where researchers and developers can test Falcon 3 prototypes before integrating them into applications.