Kimi K3
Moonshot AI · China ·
Commercial use: Conditional — commercial use with a trigger
Custom Moonshot licence, tagged "other" on the Hub — the whole Kimi line has moved off MIT. Read the repository terms before any commercial use; the earlier K2 modified-MIT attribution trigger is not a safe assumption for K3.
A trillion-parameter model built specifically to use tools and complete tasks rather than to hold a conversation.
Why it should matter to you
Relevant as evidence of where agent capability now sits, even though you will not self-host it. The licence has an unusual clause worth a note in your compliance file: above 100 million monthly users or 20 million dollars monthly revenue you must display 'Kimi K2' in your interface. Almost nobody triggers it; it is still the kind of term that should not surprise you in year three.
How it connects to our work
Agentic capability in a model does not remove the need for the surrounding system. A model that can call tools still needs guardrails, observability, and a defined blast radius, because the failure mode of an agent that acts is materially worse than one that only talks.
From our field notesGuardrails and observability are not optional once the model can act.
Advantages
- Built specifically for agentic tool use rather than chat, and it shows on tool-calling work
- 2.4M downloads within days of release
- Long context that holds up in agent loops
Disadvantages
- Custom licence, not MIT — the family changed terms and many write-ups have not caught up
- Hosted-only for practical purposes
- Chinese infrastructure on the hosted API
Reach for it when
Evaluating what open agentic capability now looks like, and hosted agent workloads where you have read and accepted the licence.
Where it falls down
Any commercial deployment where someone assumed "Kimi is MIT" from older documentation. That was K2. Verify K3 for yourself.
Running it
Hosted only, realistically. See the hardware sizing tables for how that maps to specific chips and cards, and the quantisation guide for what you give up at each bit width.
Free tiers carrying this model
| Provider | Type | The catch | Live limits |
|---|---|---|---|
| OpenRouter | Aggregator | Free-pool membership changes without notice — a model you built on can stop being free, and the list above will drift. Data passes through OpenRouter and then the upstream provider, so check the privacy and routing settings before sending anything sensitive. | check → |
| NVIDIA NIM | Inference host | It is a credit grant, not a permanent free tier. When the credits run out you are on a paid plan or self-hosting the container. | check → |
Jurisdiction: China
This is the crucial split most write-ups miss: using a Chinese lab's HOSTED API sends your data to Chinese infrastructure and is a real procurement question. Downloading their OPEN WEIGHTS and running them on your own hardware, or on a Western host, sends nothing anywhere. DeepSeek and Z.ai publish under MIT; Alibaba publishes most of Qwen3 under Apache 2.0. Those are among the most permissive licences on this page.
Watch for: Several US states and a number of government bodies restrict Chinese-hosted AI services on official devices. That restriction is about the hosted service, not about the weights running in your own VPC — but expect to have to explain the difference to a security reviewer.
This model has a published technical report: arXiv:2507.20534 It is indexed in our research library alongside the work it builds on.
Frequently asked
Can I use Kimi K3 commercially?
Custom Moonshot licence, tagged "other" on the Hub — the whole Kimi line has moved off MIT. Read the repository terms before any commercial use; the earlier K2 modified-MIT attribution trigger is not a safe assumption for K3.
What hardware do I need to run Kimi K3?
Hosted only, realistically. Weights alone are roughly 560 GB at Q4_K_M and 1050 GB at Q8_0. Add KV cache on top of that, which grows with your context length.
What licence is Kimi K3 released under?
A custom licence written by Moonshot AI, not a standard open-source licence. Custom Moonshot licence, tagged "other" on the Hub — the whole Kimi line has moved off MIT. Read the repository terms before any commercial use; the earlier K2 modified-MIT attribution trigger is not a safe assumption for K3.
Where can I use Kimi K3 for free?
Free tiers carrying it include OpenRouter, NVIDIA NIM. Limits differ per provider and change often, so check each provider's own limits page. You can also self-host the weights, which has no rate limit at all.
Similar models
DeepSeek-V4-Flash
Hosted, or 256 GB+ unified memory at Q4. Not a laptop model.
GLM-5.3-Flash
80-96 GB at Q4_K_M. A workstation or Ultra-class Mac.
Qwen3.8 27B
24 GB GPU at Q4_K_M, or a 32 GB Mac at Q6_K.
Mistral Large 3 675B
Multi-GPU datacentre, or hosted on La Plateforme.
Wiring Kimi K3 into something real?
We build the evaluation harness, the failover and the cost ceilings around a model like this, so it survives contact with production.
Maps to AI agent development, AI automation development and AI strategy and consulting. Or see it working: our case studies.