Moonshot AI Unveils Kimi K3, World’s Largest Open Model, To Rival OpenAI and Anthropic

China’s open-source AI push just took a major leap forward. Moonshot AI released Kimi K3, a 2.8 trillion-parameter model the Beijing-based startup reportedly calls the largest open-weight AI system in the world.

The company says K3 surpasses some of the most capable systems from OpenAI and Anthropic on select benchmarks, even as it continues to trail the very top tier.

The release follows earlier Financial Times reporting that Moonshot’s newest model was expected to rival and potentially exceed Anthropic’s Opus 4.8.

Notably, Anthropic has accused Moonshot, along with several other Chinese AI firms, of using model distillation earlier this year.

Where Kimi K3 Beats, and Where it Trails, its Rivals

Moonshot’s own results, cited by CNBC, show K3 beating Claude Opus 4.8 and GPT-5.5 on a range of benchmarks, including coding and general agent tasks, while consistently outperforming every other model the company tested against it. 

The model still trails Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol on overall performance, Moonshot acknowledged, placing K3 just below the current frontier rather than ahead of it. 

That result builds on real momentum: TechCrunch notes Moonshot’s earlier Kimi K2 models were already well received in the open-source AI market, ranking high on benchmarks and trailing the latest frontier systems by a relatively narrow margin. 

K3 was designed specifically to push that gap even tighter, and the benchmark results suggest it succeeded against everything except the very newest releases from Anthropic and OpenAI.

A Milestone for Open-Weight Models

The scale of K3 is itself notable. 

According to FT’s report, the model was expected to carry a parameter count between 2 trillion and 3 trillion, and Moonshot’s official release confirmed a figure of 2.8 trillion, making it the largest open-weight model available anywhere. 

Bank of America analysts, in a note led by Alex Liu, framed the achievement as evidence that Chinese labs can still produce major leaps in capability. 

As per CNBC, analysts credited pre-training scaling and architectural innovation for the step-change gains in flagship Chinese models. 

The achievement is notable given the hardware and computing constraints Chinese AI companies continue to face, long seen as a disadvantage compared with better-funded US rivals.

Market Context and Washington’s Response

The timing lines up with a broader shift in enterprise thinking, as growing debate over whether it’s worth paying premium prices for closed-source models from labs like OpenAI and Anthropic.

Moreover, as TechCrunch notes, some executives are instead pointing customers toward cheaper open-source alternatives such as DeepSeek, Z.ai, and Moonshot. 

The release reflects the intensifying US-China AI rivalry, with Chinese models gaining traction as they narrow the performance gap at lower operating costs than leading US models, though Microsoft now aims for this same cost-advantage pitch.

The trend has drawn the attention of US lawmakers, who CNBC reports are considering limits on the adoption of Chinese AI models by domestic companies. 

Source: Chinese startup Moonshot AI unveils Kimi model it says rivals OpenAI, Anthropic