• 2 min read
Moonshot’s Kimi K3 Takes Aim at US AI Leaders
Moonshot’s Kimi K3 is a 2.8-trillion-parameter open-weight model targeting frontier coding and reasoning performance.
Image: TechRadar
Moonshot has launched Kimi K3, a new AI model that the Chinese company says can outperform most US rivals, with only Claude Fable 5 from Anthropic and GPT-5.6 from OpenAI ranking higher in its comparisons.
According to Bloomberg, Kimi K3 has 2.8 trillion parameters. Artificial Analysis also ranked it ahead of Anthropic’s Opus 4.8 on some benchmarks. Moonshot says the model beats Chinese rival Z.AI on coding tasks, an area that has become one of the most commercially valuable parts of the AI market.
Moonshot describes Kimi K3 in a blog post as the “world’s first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.” It is available to use now.
Bloomberg quoted Leonid Mironov, a portfolio manager at Gavekal Capital, who said:
Recommended reading
Copilot Could Diagnose Your PC—If It Doesn’t Slow It Down
“In my use, it’s clearly the best Chinese model ever,” adding that it was “brilliant.”
Why Kimi K3's open weights matter
Kimi K3's biggest competitive distinction is its open-weight approach. Moonshot plans to release the model weights on July 27, allowing organizations to download and run the pretrained model without paying for access. That does not make it open source: users can obtain the model, but not inspect how it was trained or which data was used.
Running a 2.8-trillion-parameter model will require extremely powerful hardware. Still, companies with the resources to operate it will be able to use a frontier-capable pretrained system at no licensing cost, potentially putting pressure on OpenAI and Anthropic.
Moonshot is charging more than Chinese competitors for hosted access, pricing Kimi K3 at roughly Claude Sonnet levels. That price is itself a signal that the company views the model as comparable with current frontier systems, particularly for coding.
The rapid push to improve performance has also renewed concerns that AI safeguards could be overlooked in favor of faster development. Those concerns come alongside accusations of AI theft made by the US State Department against Chinese firms earlier this year, including Moonshot.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via TechRadar
