Moonshot AI Releases Kimi K3 Weights: 2.8 Trillion Parameters, Tighter Commercial License
Moonshot AI has released the weights for Kimi K3, a 2.8 trillion parameter model weighing in at 1.56TB on Hugging Face. The new license drops the 'modified MIT' framing and now requires companies earning over $20 million in 12-month revenue from Model-as-a-Service offerings to sign a separate agreement with Moonshot.
Moonshot AI has released the weights for Kimi K3, a 2.8 trillion parameter model available now on Hugging Face. The release follows an announcement Moonshot made a few weeks earlier previewing the model. The full weights total 1.56TB.
The release comes with a significant licensing change. Kimi's K2 model, released in January, used a modified version of the MIT license that required companies with more than 100 million monthly active users, or more than $20 million in monthly revenue, to display "Kimi K2" prominently in their product's user interface.
K3 goes further. The new license no longer describes itself as "modified MIT." Instead, it states that any Model-as-a-Service business whose aggregate revenue — including affiliates — exceeds $20 million over any consecutive 12-month period must enter into a separate commercial agreement with Moonshot AI before using the software or any derivative works commercially.
Moonshot does not market K3 as "open source." The company consistently uses the term "open weight" in its own materials, an accurate description given the licensing restrictions now attached to large-scale commercial use.
Availability and pricing
OpenRouter is already listing K3 from seven providers. Most offer pricing matching Moonshot AI's own rates: $3 per million input tokens and $15 per million output tokens. Moonshot has not disclosed context window size or published benchmark scores alongside this release; those details are not yet confirmed.
What this means
K3's license marks a clear escalation from K2's attribution requirement to an outright gating mechanism for large commercial deployments. Any company running a Model-as-a-Service business past the $20 million revenue threshold — over any rolling 12 months, not just monthly — now needs Moonshot's direct sign-off to keep using the weights commercially. That's a meaningfully higher bar of control than a UI attribution badge, and it puts Kimi's licensing closer to a dual-license commercial model than to anything resembling permissive open source.
For most developers and startups experimenting with K3, none of this matters — the $20 million revenue threshold is high enough that the vast majority of users will never trigger it. But for larger AI infrastructure providers and API resellers, the calculus changes: building a business on top of K3 weights now carries a legal dependency on Moonshot's continued willingness to negotiate.
The pricing itself remains competitive. At $3/$15 per million tokens, K3 sits in the same tier as several other frontier-adjacent Chinese models now available through OpenRouter, continuing a pattern where Chinese labs are shipping large, capable models at prices well below the leading Western labs' flagship offerings — even as the licensing terms attached to those weights grow more restrictive with each release.
Related Articles
Moonshot AI Releases Kimi K3: 2.8T-Parameter Open-Weight Model with 1M-Token Context, Now Available via Unsloth Quantiza
Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight mixture-of-experts model with a 1-million-token context window and native multimodal support. Unsloth has published Dynamic 2.0 quantized versions on Hugging Face, claiming improved accuracy over other quantization methods.
Moonshot AI Open-Sources Kimi K3 Weights After Model Matched GPT-5.6 Sol on Benchmarks
Moonshot AI has released open weights, a technical report, and supporting infrastructure for Kimi K3, a model that claims 2.5x more intelligence per unit of compute. Independent testing found notable gaps in cybersecurity and math performance compared to Western frontier models.
Moonshot AI Releases Kimi K3: Open-Weight 2.8T-Parameter Model With 1M-Token Context and Native Multimodality
Moonshot AI has released Kimi K3, an open-weight 2.8-trillion-parameter mixture-of-experts model with 104B activated parameters, a 1,048,576-token context window, and native multimodal support. The company describes it as the world's first open 3T-class model, built on a new Kimi Delta Attention architecture.
Alibaba Launches Qwen3.7 Flash: 1M-Context Vision-Language Model at $0.03/$0.13 per 1M Tokens
Alibaba has released Qwen3.7 Flash, a vision-language reasoning model with a 1 million token context window aimed at multimodal agents, visual coding, and computer-use tasks. The model is priced at $0.03 per 1M input tokens and $0.13 per 1M output tokens and is available through OpenRouter.
Comments
Loading...