In the rapidly evolving landscape of artificial intelligence, a new contender has emerged, challenging the dominance of established models like ChatGPT and Claude. Kimi AI, developed by Chinese AI company Moonshot AI, has garnered attention for its performance on major benchmarks, suggesting a potential shift in the global AI landscape.

Kimi AI is a series of large language models, with the first version released in October 2023. It was noted for supporting a large context window of 128,000 tokens. Moonshot AI was founded in March 2023 with the goal of achieving Artificial General Intelligence (AGI) through foundational models, according to CEO Yang. Yang's milestones include long context length, multimodal world model, and a scalable general architecture capable of continuous self-improvement without human input.

Kimi's Capabilities and Features

Kimi AI distinguishes itself with several key features:

• Long Context Processing: Kimi can handle extensive documents, including entire books, lengthy legal files, and large codebases, without losing coherence. Kimi's long-context capability combined with open access shifts the market dynamic, where most high-performing models remain closed or tightly controlled.
• Multimodal Capabilities: Kimi AI can understand various input types, including text, images, files, and code. Kimi 1.5 is designed to handle both text and visual information, which allows it to analyze visual content and interpret complex datasets.
• Agentic AI Feature: In September 2025, Moonshot AI added an agentic AI feature to Kimi, known as "OK Computer". It is capable of creating multi-page websites and editable slides from simple user prompts and can process up to 1 million rows of input data at once, and output text, audio, images, and video.
• Coding Proficiency: Kimi K2 achieved state-of-the-art performance in coding benchmarks. Kimi-Dev, a 72B parameter coding-focused model based on Qwen2.5-72B, achieved state-of-the-art performance among open source models on the SWE-bench Verified benchmark.
• Reasoning Abilities: Kimi K2 Thinking can execute hundreds of sequential tool calls without human interference, reasoning coherently across steps to solve complex problems.

Benchmark Performance

The Kimi K2 Thinking model, released in November 2025, has demonstrated strong performance on several benchmarks, outperforming GPT-5 and Claude Sonnet 4.5 in some instances.

• Humanity's Last Exam (HLE): Kimi K2 Thinking achieved a score of 44.9% compared to GPT-5's 41.7% and Claude Sonnet 4.5 Thinking's 32%.
• BrowseComp: Kimi K2 Thinking outperformed GPT-5 and Claude Sonnet 4.5 Thinking with a score of 60.2%. This benchmark evaluates AI chatbots' web browsing proficiency and information-seeking persistence.
• SWE-bench Verified: Kimi K2 Thinking achieved a score of 71.3%.

However, Kimi K2 Thinking seems to lag behind ChatGPT in coding-related tasks, scoring 83.1% on LiveCodeBench V6, while GPT-5 scored 87%.

Open Source and Cost-Effectiveness

Kimi K2 Thinking is open source, allowing developers to access and build upon the underlying code and weights for free. Moonshot AI also stated that training Kimi K2 Thinking cost less than $5 million, a relatively small amount compared to the billions spent by leading AI labs in the U.S. Kimi K2's aggressive API pricing and open-source model make it a cost-effective option, especially for projects with tight budgets.

Implications

The emergence of Kimi AI highlights the increasing competition in the AI field, with Chinese models closing the gap with their Western counterparts. The open-source nature of Kimi K2 Thinking could further accelerate AI development and innovation. Kimi AI's capabilities in long context processing, multimodal understanding, and agentic tasks position it as a versatile tool for various applications, including research, education, healthcare, and creative content generation.