Tom's Hardware Article
https://www.tomshardware.com/tech-industry/artificial-intelligence/moonshot-ai-releases-weights-for-kimi-k3-firing-a-shot-across-the-bow-of-openai-and-anthropic-open-weight-model-performs-almost-as-well-as-frontier-models-while-being-2-3x-easier-to-run
Here's a summary.
Chinese company Moonshot AI has released the model weights for Kimi K3 for free.
(There is a license that requires a separate contract if a certain revenue amount is exceeded.)
https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE
Regarding inference cost comparisons, Moonshot states that its costs are "internally measured figures" compared to other companies' published token prices. However, the results are quite impressive. Kimi K3 charges $3 per million tokens. In contrast, Anthropic's Fable charges $10 per million tokens and OpenAI's GPT-5.6 Sol charges $5 per 100 thousand tokens.
This is particularly competitive, especially considering that it doesn't apply caching for general input criteria. Kimi K3's caching structure achieves a 90% hit rate in coding tasks. Therefore, if the usage environment has high cache utilization, the cost of 1 million tokens can be reduced from $3 to $0.30.
Interestingly, Moonshot's report only mentions that Kimi was run using NVIDIA's H20 for some coding tests. This is a relatively low-end chip by today's standards. Unlike the Blackwell B-series chips subject to export regulations, this GPU does not natively support MX floating-point format at the hardware level.
The fact that anyone with a decent AI GPU can become a direct competitor is bad news for OpenAI and Anthropic. Moreover, the fact that Kimi K3 is open weight (weights are released) gives the impression that "free" software can perform almost as well as proprietary competitors while being significantly cheaper to run. However, it's important to note that open weight doesn't mean open source. The learning process and dataset remain Moonshot's unique secret sauce.
[The rest of the content is omitted]
Artificial Analysis Twitter (X) post:
"Today, with the release of Kimi K3, open-weight AI has taken another step forward. The gap between leading proprietary models and open-weight models is now only 4 points according to the Artificial Analysis Intelligence Index, which is the smallest difference since GLM-5 was released in February.
Moonshot's Kimi K3 scored 57 points on the intelligence index. Only two research institutions have models with higher scores. Anthropic, which owns Claude Opus 5 (61 points) and Claude Fable 5 (60 points), and OpenAI, which owns GPT-5.6 Sol (59 points).
https://x.com/ArtificialAnlys/status/2081926991788626011
