Moonshot Unveils Kimi K3: A 2.8 Trillion Parameter Open AI Model Challenging Western Giants

The global artificial intelligence landscape is witnessing a significant shift as Chinese developers continue to push the boundaries of open-source technology. In a major move, AI startup Moonshot has officially introduced Kimi K3, a massive open model designed to compete directly with some of the industry’s most prominent proprietary systems.

Boasting an unprecedented 2.8 trillion parameters and a 1 million token context window, Kimi K3 represents a massive leap forward in scale and capability. According to its developers, the model will have its weights publicly released on July 27, allowing researchers and developers worldwide to run, analyze, and modify the system independently. This open-source approach stands in stark contrast to the closed ecosystems of many Western tech giants.

Performance and Benchmarks

In benchmark evaluations, Kimi K3 has demonstrated remarkable performance, particularly in software development and autonomous agent tasks. Moonshot reports that the model outperforms Claude Opus 4.8 and GPT-5.5 across several key industry tests, including DeepSWE, FrontierSWE, Program Bench, GPDval-AA V2 ELO, and JobBench.

While it slightly trails behind ultra-premium models like Fable 5 and GPT-5.6 Sol, Kimi K3 remains highly competitive. Beyond standard coding tasks, the model features native multimodal capabilities. It leverages advanced vision, programming, and 3D reasoning to generate interactive games from either code or static images. Additionally, it can analyze complex datasets to produce interactive charts and presentations.

A Massive Leap in Scale

The sheer size of Kimi K3 marks a dramatic evolution in open-source AI. Its predecessor, Kimi K2, featured just over 1 trillion parameters. By scaling up to 2.8 trillion parameters, Moonshot has established a new benchmark for open models. For comparison, other leading Chinese open-weight models include DeepSeek V4 Pro at 1.6 trillion parameters and Xiaomi’s MiMo V2.5 Pro at 1.02 trillion parameters.

This rapid expansion highlights the intense competition within the Chinese AI sector, where companies are aggressively scaling their architectures to match or exceed the capabilities of US-based labs.

Disruptive Pricing Strategy

In addition to its technical capabilities, Moonshot is positioning Kimi K3 as a highly economical alternative for developers accessing the model via its API. The pricing structure is set at $3 per million input tokens and $15 per million output tokens.

This makes Kimi K3 significantly more affordable than its direct Western competitors. For instance, Claude Opus 4.8 costs $5 per million input tokens and $25 per million output tokens. This aggressive pricing could accelerate adoption among startups and enterprise clients looking to manage operational costs without sacrificing performance.

To ensure immediate utility, Moonshot has already integrated the new model into its existing product ecosystem, including Kimi Code, Kimi Work, and the official Kimi mobile application.