Kimi K3.1 Pro
Kimi K3.1 Pro is an open-weight, natively multimodal agentic model and the first model in the K3.1 series. It is post-trained from Kimi K3. Its mixture-of-experts architecture has 2.8 trillion total…
MODEL DIRECTORY
New releases, in one place. Explore models by their publication date.
Kimi K3.1 Pro is an open-weight, natively multimodal agentic model and the first model in the K3.1 series. It is post-trained from Kimi K3. Its mixture-of-experts architecture has 2.8 trillion total…
Kimi K3.1 Pro Highspeed is the faster-serving version of Kimi K3.1 Pro, a 3-trillion-parameter-class, open-weight model developed through additional post-training of Kimi K3. Moonshot AI says…
Kimi K3.1 Pro is the first model in the K3.1 series and an open-weight, multimodal model built by post-training Kimi K3. Moonshot positions it for complex coding, visual understanding, research, and…
K2.8 Preview is a preview coding and agent model offered through Kimi Code. It was fully rolled out on September 11, 2026, under the existing kimi-for-coding model ID. Moonshot describes its overall…
Kimi K3 is an open-weight, natively vision-capable model designed for long-horizon coding, knowledge work, and reasoning. It has 2.8 trillion total parameters and a mixture-of-experts architecture…
Kimi K2.7 Code is an open-weight, coding-focused model designed for long-horizon software engineering and agent workflows. It was released on June 12, 2026. (kimi.com) Supports text, image and video…
Kimi K2.6 is an open-weight, native multimodal model designed for general-purpose and agentic tasks, with an emphasis on long-horizon coding, interface design and autonomous workflows. It accepts text…
Kimi K2.5 is a native multimodal model designed for visual understanding, coding and agentic tasks. It was released on January 27, 2026. Moonshot AI built it on Kimi K2 through continued pretraining…
Kimi K2 Thinking is a reasoning model designed for agentic tasks. It interleaves step-by-step reasoning with tool calls for research, coding, writing, and complex problem-solving. Moonshot AI…
Kimi-Linear-48B-A3B-Instruct is an instruction-tuned, text-generation model designed in part for efficient long-context processing. It has 48 billion total parameters, approximately 3 billion…
Kimi-Linear-48B-A3B-Base is the pretrained base checkpoint of Moonshot AI’s Kimi Linear language-model family. It has 48 billion total parameters, approximately 3 billion activated parameters, and a…
Kimi-K2-Instruct-0905 is an instruction-tuned Kimi K2 language model aimed at coding and agentic tasks. It uses a mixture-of-experts architecture with 1 trillion total parameters and 32 billion…
Kimi-K2-Instruct is a post-trained instruction model in the Kimi K2 family, intended for general conversation and agent applications, with a focus on tool calling and task execution rather than a long…
Kimi-K2-Base is a pretrained base language model in the Kimi K2 series, primarily intended for researchers and developers who want to fine-tune it themselves and build custom solutions. It differs…
Kimi-VL-A3B-Thinking-2506 is a multimodal vision-language reasoning model in the Kimi-VL series, intended for image understanding, long-context, and agent tasks. It uses a mixture-of-experts (MoE)…
Kimi-Researcher is an autonomous agent model for complex research tasks, based on an internal Kimi k-series model and trained through end-to-end agent reinforcement learning. The official announcement…
Kimi-Dev-72B is an open-source code large language model for software engineering problem solving. It is based on Qwen2.5-72B and has been trained on code-related data and optimized through…
kimi-thinking-preview is a preview version of a long-thinking model released by Moonshot AI. The official description presents it as a model with multimodal and general reasoning capabilities that…
Kimi-Audio-7B-Instruct is an instruction-tuned version of the Kimi-Audio audio foundation model, designed for audio understanding, generation, and dialogue. The model weights and inference code for…
Kimi-VL-A3B-Thinking is a vision-language reasoning model in the Kimi-VL series, focused on complex text and multimodal reasoning. It uses a mixture-of-experts (MoE) architecture, with approximately 1…