Kimi or DeepSeek: large open models for agentic work
Both target GPUs and clusters, both offer context up to 1M tokens and both understand images in certain versions. DeepSeek is simpler on licensing and setup: recent versions are MIT, the family is in Ollama and the smallest size is 7B. Kimi starts at 16B-A3B, is absent from Ollama, and its terms vary by version: K2 through K2.7-Code use a modified MIT that requires crediting Kimi in the interface at 100 million users or 20 million dollars of monthly revenue, while K3 has its own license under which API businesses above 20 million dollars a year need a separate agreement. In return, K3 at 2.8 trillion parameters was the largest open model at release, and K2.7-Code 1T-A32B is built for programming.
Comparison based on catalog data
| Parameter | Kimi | DeepSeek |
|---|---|---|
| Category | Text, Image + text, Code | Text, Image + text |
| Developer | Moonshot AI, China | DeepSeek, China |
| Releases | Feb 2025 – Jul 2026 | Nov 2023 – Sep 2026 |
| Sizes | 16B-A3B – 2.8T-A104B | 7B – 1.6T-A49B |
| Hardware | 1 GPU, Cluster | Laptop, 1 GPU, Cluster |
| Commercial use | Commercial use with conditions | Commercial use allowed |
| License | K2–K2.7-Code: modified MIT (at 100 million users or $20 million monthly revenue, Kimi must be credited in the interface); K3: own Kimi K3 License, API businesses with revenue from $20 million a year need a separate agreement | MIT (V2 and earlier versions: DeepSeek's own license prohibiting harmful use) |
| Russian | Not stated | Not stated |
| Ollama | No | Yes |
| Without GPU | No | No |
| Tasks |
|
|
Choose Kimi if
- You need multi-step agents: search, data collection, reporting
- You have a cluster that can host a 2.8 trillion parameter model
- You want a dedicated coding model: K2.7-Code 1T-A32B
Choose DeepSeek if
- You want MIT without revenue thresholds or attribution requirements
- You install with a single Ollama command
- You are starting small: DeepSeek begins at 7B
Other comparisons
- DeepSeek or Qwen: which to choose for business
- Llama or DeepSeek: which to run inside your own perimeter
- Qwen or GigaChat: which to choose for business
- Qwen or Llama: which to choose for business
- GigaChat or YandexGPT: which to choose for business
- Gemma or Llama: which to choose for business
- Mistral or Llama: which to choose for business
- gpt-oss or Qwen: which one to run on your own server
- Mistral or Qwen: which to choose for business
- Gemma or Phi: small models for modest hardware
- YandexGPT or Qwen: which to choose for Russian
- T-Pro or GigaChat: which one for a Russian company
Need a model for your task?
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
- SelectThe model and size for your task and hardware budget
- DeployOn your server or in a closed network, with an API
- Fine-tuneOn your data, or connect a knowledge base
- IntegrateInto your CRM, ERP, bot, website or team chat