China's Moonshot AI unveils Kimi K3, claims world's first 3-trillion-parameter open-source model

Chinese AI startup Moonshot AI has unveiled Kimi K3, an AI model it says rivals the performance of leading American AI systems.

Moonshot AI said Saturday it had released Kimi K3, a model with about 2.8 trillion parameters that combines native vision capabilities with a 1-million-token context window — the span of text a language model can reference when generating a response.

The company said Kimi K3 is the world's first open-source model at the "3-trillion-parameter scale," designed for advanced intelligence scenarios including long-horizon coding tasks, intellectual work and reasoning.

Industry observers say the model matches the performance of major American AI systems backed by massive capital investment. In fact, AI performance research firm Artificial Analysis gave Kimi K3 an intelligence score of 57, placing it fourth among all existing AI models.

That puts it ahead of xAI's Grok 4.5 (54), Meta's Llama Spark 1.1 (51) and Google's Gemini 3.5 Flash (50). It trails Anthropic's Claude Fable 5 (60), OpenAI's GPT-5.6 Sol Max (59) and GPT-5.6 Sol X-High (58) by a narrow margin.

Moonshot AI said Kimi K3 is built for long-horizon tasks, specializing in processing large software repositories, operating terminal tools, and combining coding with visual feedback rather than simple question-and-answer exchanges. The model can also modify digital content such as interfaces and games through its image and video understanding capabilities, the company said.

China Daily and other outlets noted that Kimi K3 departs from the low-cost strategy Chinese AI models have typically pursued.

The model's API is priced at 100 yuan ($15) per 1 million output tokens — higher than DeepSeek V4 Pro and Xiaomi's MiMo-V2.5 Pro (6 yuan each), Z.ai's GLM-5.2 (28 yuan) and Alibaba's Qwen 3.7 Max (36 yuan).

Meanwhile, following the "DeepSeek shock" that reverberated globally early last year, international observers are closely watching the emergence and potential impact of high-performance models like Kimi K3. Morgan Stanley said the release of Kimi K3 shows that Chinese large language models are catching up with leading American firms in scale, performance and pricing.