Just days after DeepSeek, another Chinese AI company Moonshot launches model Kimi k1.5 that outshines OpenAI

The AI arms race has taken another turn as China’s Moonshot AI has launched its latest model, Kimi k1.5, just days after the rise of DeepSeek’s DeepSeek-R1. While DeepSeek has already been seen as a formidable competitor to OpenAI’s GPT-4, Kimi k1.5 is now being touted as an even more powerful alternative, outshining OpenAI’s GPT-4o and Claude 3.5 Sonnet on key benchmarks.  

This move marks another significant milestone in China’s growing influence in the AI space, challenging the dominance of US tech giants.

What is Kimi k1.5?

Kimi k1.5, developed by Beijing-based startup Moonshot AI, is a multimodal AI model that integrates visual, text, and code inputs to solve complex problems. It is being hailed as a direct competitor to OpenAI’s GPT-4o, and some reports suggest that it even outperforms GPT-4 in areas such as mathematics, coding, and understanding text and visual data. Unlike DeepSeek’s DeepSeek-R1, which lacks multimodal capabilities, Kimi can process and reason across text and images, giving it an edge in tasks requiring both formats. What makes Kimi particularly impressive is that it was built at a fraction of the cost it would take to develop a similar frontier AI in the US. According to reports, Kimi k1.5 has been dubbed the first true rival to OpenAI’s models.

Kimi’s unique features

Kimi k1.5 isn’t just another AI model; it represents a significant leap in reinforcement learning (RL) and multimodal reasoning. The model uses RL techniques to enhance its decision-making process by rewarding itself through exploration. This allows Kimi to break down complex problems into manageable steps, improving its reasoning ability. Kimi is designed to handle long-context tasks, processing up to 128,000 tokens, which allows it to understand and generate responses based on vast amounts of data. Its ability to combine visual data, text, and code makes it highly versatile and well-suited for a wide range of applications.

In terms of benchmarks, Kimi has outperformed both GPT-4o and Claude 3.5 Sonnet in several areas. It scored 96.2 on MATH 500, outperforming GPT-4’s variants, and 77.5 on AIME, a math benchmark. It also scored in the 94th percentile on Codeforces, a competitive coding platform. The model has surpassed its US counterparts by up to 550% on some benchmarks, particularly in areas like problem-solving and reasoning. However, the reliability of these scores is often questioned, as AI companies typically conduct their own benchmark tests and publish results.

More from Tech

How ChatGPT is becoming everyone’s BFF and why that’s dangerous America ready for self-driving cars, but it has a legal problem

Kimi’s impact

Kimi k1.5’s efficiency and versatility have set it apart from many existing AI models, and its introduction has raised eyebrows in the AI community. As it continues to outperform leading US models in areas such as reasoning, math, and long-context tasks, Kimi may be poised to revolutionize industries that rely heavily on AI, such as healthcare, engineering, and data analysis. Despite questions over the reliability of benchmark tests, the impact of Kimi’s advancements is undeniable.

With its multimodal capabilities and reinforcement learning approach, Kimi k1.5 is expected to play a significant role in shaping the future of AI. As China strengthens its foothold in the AI race, Kimi’s rise is a clear signal to US tech companies that they must innovate faster to keep up with this rapidly advancing field.