|
The rise of Chinese AI emerges as one of the key factors that drove the Artificial Intelligence landscape in the first quarter of 2025, according to a report by Artificial Analysis, an independent AI benchmarking and insights provider. The US leads across reasoning models, the top 4 places on the Artificial Analysis Intelligence Index are all taken by reasoning models from US labs. However, China currently leads non-reasoning models; DeepSeek V3 0324 is the leading non-reasoning model. Models from labs outside of USA and China have continually improved, but currently don't compete for frontier intelligence.
The rise of Chinese AI Labs marked a notable shift in global competition. Chinese entities released models that rivalled US counterparts, particularly in open-weight models, with DeepSeek V3 0324 leading in non-reasoning categories. The report also highlights that AI continues to progress, with the help of leading labs like OpenAI, Google, and many others, consistently advancing intelligence, efficiency, and speed. While OpenAI's o4-mini (high) maintained a lead, models like Google's Gemini 2.5 Pro and xAI's Grok 3 rapidly narrowed the gap, intensifying competition. In the field of AI, reasoning models emerged as a significant frontier. These models, which "think" through problems by generating intermediate steps before providing an answer, became widespread across major labs. This approach has delivered substantial intelligence gains, influencing how AI processes complex tasks. Artificial Analysis also discusses the efficiency and Mixture of Experts (MoEs) in revolutionising AI inference. Costs plummeted by over 32 times since September 2024, and more than 1000x since GPT-4's 2023 launch. This was primarily due to smaller models (including MoE architectures), optimised inference, and new hardware, making advanced AI more accessible Lastly, Multimodal AI saw significant advancements. Models became more capable of handling image and audio data natively. Image generation, exemplified by OpenAI's GPT-4o, achieved new quality benchmarks, while text-to-speech models improved significantly, offering more human-like dialogue. These trends collectively define a quarter of rapid evolution, increased accessibility, and a globally competitive environment in AI. (ANI)
|
