Alibaba has updated its artificial intelligence (AI) model Qwen and claimed that the new release Qwen 2.5-Max outperforms DeepSeek-V3, Reuters reports.
The latest Alibaba technical report on Qwen is the "Qwen2 Technical report" published on arXiv in September 2024.
Qwen2.5-Max "has been pretrained on over 20 trillion tokens and further post-trained with curated Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF) methodologies," says Alibaba's announcement. "Today, we are excited to share the performance results of Qwen2.5-Max and announce the availability of its API through Alibaba Cloud. We also invite you to explore Qwen2.5-Max on Qwen Chat!"
The announcement presents the performance results of Qwen2.5-Max alongside leading state-of-the-art models, including DeepSeek V3, GPT-4o, and Claude-3.5-Sonnet.
Chinese AI technology in rapid motion
DeepSeek's AI models, V3 and the recently released R1, have shaken up the tech world, causing stock prices to drop. Investors now question the high costs of AI development in the US due to DeepSeek's low-cost strategy.
As reported by Reuters, DeepSeek's founder Liang Wenfeng said that his company focuses on AGI, or artificial general intelligence. Liang believes big tech companies might not be ideal for future AI due to their high costs and rigid structures. He thinks smaller, more flexible teams like DeepSeek can innovate better. "Large foundational models require continued innovation, tech giants' capabilities have their limits," he said.
DeepSeek's achievements have pushed Chinese competitors like Alibaba to improve their AI. DeepSeek-V2, the predecessor of V3, started an AI price war in China by being very cheap and open-source. This led Alibaba to slash prices on its models by up to 97%. Other companies like Baidu and Tencent also cut costs following suit.
Reuters also reports that TikTok owner ByteDance quickly updated its AI model after DeepSeek-R1's release, claiming it beat OpenAI's o1 in a benchmark test known as AIME, which checks how well AI understands complex instructions. DeepSeek-R1 also claimed to match OpenAI's o1 in several benchmarks.
