Free Chinese AI claims to beat GPT-5

Kimi K2 Thinking: A New Contender

Moonshot introduced Kimi K2 Thinking on Thursday, positioning it as a powerful reasoning model. The company asserts that Kimi K2 Thinking outperforms OpenAI's GPT-5 and Anthropic's Claude Sonnet 4.5 in several critical areas. These include:

1. Humanity's Last Exam: A benchmark assessing general knowledge and reasoning.
2. BrowseComp: Evaluates an AI agent's ability to extract complex information from the internet.
3. Seal-0: Measures advanced reasoning capabilities.

While Kimi K2 Thinking demonstrated comparable coding abilities to GPT-5 and Sonnet 4.5, its primary strength lies in its reasoning and adaptive capabilities. Moonshot describes the model's approach on its website:

By reasoning while actively using a diverse set of tools, K2 Thinking is capable of planning, reasoning, executing, and adapting across hundreds of steps to tackle some of the most challenging academic and analytical problems.

Model Architecture and Open-Source Advantage

Kimi K2 Thinking is built as a Mixture-of-Experts (MoE) model, integrating long-horizon planning with adaptive reasoning. It incorporates the use of online tools, such as web browsers, enabling it to:

1. Continuously generate and refine hypotheses.
2. Verify evidence and construct coherent answers.
3. Decompose complex, ambiguous problems into clear, actionable subtasks.

The model was trained with approximately 1 trillion parameters and is accessible via Hugging Face. A significant aspect of Kimi K2 Thinking, which builds on the earlier Kimi K2 model released in July, is its open-source availability. This means developers can access and utilise its underlying code and weights without charge.

This open-source approach, combined with Moonshot's claim of superior agentic capabilities compared to proprietary models, presents a compelling proposition. Furthermore, Moonshot stated the training cost for Kimi K2 Thinking was less than $5 million – specifically $4.6 million, according to CNBC. This figure is notably small when compared to the billions invested by prominent AI laboratories in the United States. Should these performance claims be independently verified, the implications for the AI industry could be substantial.

Implications for Businesses

The rapid advancements in AI, particularly the emergence of sophisticated AI agents, have created pressure for businesses to integrate these tools. Since the launch of ChatGPT nearly three years ago, companies have been encouraged to adopt AI solutions, often marketed as productivity enhancements and virtual assistants. This typically involves investing in enterprise-level offerings, such as OpenAI's ChatGPT for Enterprise.

The arrival of an open-source model like Kimi K2 Thinking, which purports to offer advanced capabilities at a significantly lower training cost, could disrupt existing market dynamics. It might provide businesses with more accessible and cost-effective alternatives for deploying AI agents, potentially challenging the dominance of proprietary, high-cost solutions. For instance, smaller businesses might find new ways to survive Google's AI Overview by leveraging such open-source models. This shift could also impact how companies approach their generative AI adoption strategies. Further research into open-source AI models can be found in publications like those from the MIT Technology Review.