Fine-tuning and research on speech recognition, audio understanding, audio classification, audio question answering, and speech-audio generation
Type
Multimodal
Context
8K
Reasoning
4/10
Speed
5/10
Multimodal
Audio input
Media output
Status
Current open-weight base model; not instruction-tuned
Input
No official hosted API pricing; downloadable open weights
Output
No official hosted API pricing; downloadable open weights
View model
→
Self-hosted speech recognition, audio understanding, audio question answering, audio captioning, and spoken conversational agents
Type
Multimodal
Reasoning
6/10
Speed
5/10
Multimodal
Audio input
Media output
Status
Available as an open-weight self-hosted model
Input
No official hosted API pricing
Output
No official hosted API pricing
View model
→
Repository-level software issue resolution, code repair, test generation, coding agents, and self-hosted software engineering workflows
Type
Coding
Context
131K
Reasoning
7/10
Speed
4/10
Status
Available open-weight model
View model
→
Self-hosted image and video understanding, OCR, long-document analysis, visual question answering, screenshot perception, and efficient multimodal applications
Type
Multimodal
Context
131K
Reasoning
7/10
Speed
8/10
Multimodal
Image input
Video input
Status
Current open-weight/downloadable model; a newer Kimi-VL-A3B-Thinking-2506 variant is recommended for stronger multimodal reasoning
Input
No official Moonshot-hosted API price published; self-hosted/open-weight model
Output
No official Moonshot-hosted API price published; self-hosted/open-weight model
View model
→
Local multimodal reasoning, mathematical visual question answering, OCR, document understanding, image and video analysis, and research on open-weight vision-language models
Type
Reasoning
Context
131K
Reasoning
7/10
Speed
7/10
Multimodal
Image input
Video input
Status
Available as downloadable open weights; superseded by Kimi-VL-A3B-Thinking-2506
View model
→
Open-weight image and video reasoning, OCR, chart interpretation, visual mathematics, long PDFs, high-resolution screenshots and GUI-agent grounding
Type
Reasoning
Context
131K
Reasoning
8/10
Speed
7/10
Multimodal
Image input
Video input
Status
Current open-weight model; publicly available for self-hosted and third-party inference
Input
No official hosted API price; open-weight model
Output
No official hosted API price; open-weight model
View model
→
Fine-tuning, foundation-model research, custom language systems, coding experiments, and self-hosted inference
Type
General Purpose
Context
131K
Reasoning
8/10
Speed
3/10
Fine-tuning
Streaming
Status
Open-weight and downloadable; legacy relative to newer Kimi K2.x models but still accessible from the official Hugging Face repository
Input
No official hosted API price for the Base checkpoint; self-hosted weights
Output
No official hosted API price for the Base checkpoint; self-hosted weights
View model
→
Open-weight coding assistants, tool-using agents, general-purpose chat, and self-hosted research deployments
Type
General Purpose
Context
131K
Reasoning
8/10
Speed
6/10
Tool use
Streaming
Status
Open-weight checkpoint available; legacy relative to Moonshot AI's current hosted API catalog
Input
USD 0.60 per 1 million input tokens historically for the Kimi K2 preview API; current first-party pricing for this exact model is unverified
Output
USD 2.50 per 1 million output tokens historically for the Kimi K2 preview API; current first-party pricing for this exact model is unverified
View model
→
Self-hosted reasoning agents, autonomous research, long-horizon tool workflows, coding, and complex multi-step analysis
Type
Reasoning
Context
262K
Reasoning
9/10
Speed
5/10
Tool use
Streaming
Status
Retired from Moonshot's direct API; open-weight checkpoint remains available for self-hosted or third-party deployment
View model
→
Long-horizon software engineering, agentic coding, visual document understanding, tool-using workflows, and multi-agent orchestration
Type
Multimodal
Context
262K
Reasoning
8/10
Speed
8/10
Multimodal
Image input
Tool use
Status
Current; open-source model, available through Kimi, Kimi API, Kimi Code, and downloadable model weights
View model
→
Long-horizon software engineering, repository-level coding, multi-file refactoring, debugging, coding agents, and tool-driven development workflows
Type
Coding
Context
262K
Reasoning
8/10
Speed
7/10
Multimodal
Image input
Video input
Status
Current and accessible through Kimi API; Kimi Code default service has moved to Kimi K2.8 Preview, while Kimi K2.7 Code remains available through API and Kimi K2.7 Code HighSpeed service.
Input
¥6.50 per 1M uncached input tokens; ¥1.30 per 1M cache-hit input tokens
Output
Â¥27.00 per 1M output tokens
View model
→
Long-context software development, repository analysis, code completion, multi-file refactoring, and agentic coding workflows
Type
Coding
Context
1.05M
Reasoning
8/10
Speed
7/10
Multimodal
Image input
Video input
Status
Preview; fully rolled out in Kimi Code
Input
Not publicly listed; accessed through Kimi Code membership and quota
Output
Not publicly listed; accessed through Kimi Code membership and quota
View model
→
Multimodal coding, visual debugging, long-context analysis, tool-using agents, and complex research or office workflows
Type
Multimodal
Context
256K
Reasoning
9/10
Speed
8/10
Multimodal
Image input
Video input
Input
$0.60 per 1 million tokens; cached input approximately $0.10 per 1 million tokens
Output
$3.00 per 1 million tokens
View model
→
Long-context coding, software engineering, multimodal document and video understanding, agentic workflows, technical research, and complex reasoning
Type
Multimodal
Context
1M
Reasoning
9/10
Speed
5/10
Multimodal
Image input
Video input
Status
Current and available; open-weight model
Input
$3.00 per MTok cache-miss input; $0.30 per MTok cache-hit input
Output
$15.00 per MTok
View model
→
Long-context text generation, local inference, continued pretraining, research, and fine-tuning
Type
Lightweight
Context
1.05M
Reasoning
7/10
Speed
9/10
Fine-tuning
Streaming
Status
Current open-weight model; publicly downloadable
Input
Not applicable; open-weight checkpoint with no official hosted API price
Output
Not applicable; open-weight checkpoint with no official hosted API price
View model
→
Long-context text generation, self-hosted assistants, document analysis, research, and efficient local inference
Type
General Purpose
Context
1.05M
Reasoning
7/10
Speed
9/10
Streaming
Status
Current open-weight instruction-tuned checkpoint
View model
→
Self-hosted text generation, language-model research, efficient MoE inference, code and mathematics experimentation, and fine-tuning research
Type
General Purpose
Context
8K
Reasoning
6/10
Speed
8/10
Status
Available open-weight pretrained checkpoint
View model
→
Self-hosted instruction following, general text generation, research, experimentation, and cost-conscious local inference
Type
General Purpose
Context
8K
Reasoning
6/10
Speed
8/10
Status
Available open-weight checkpoint
View model
→
Native-resolution image feature extraction, vision-language model backbones, high-resolution document and image understanding, and multimodal research
Image input
Status
Current open-weight model; available for local use through Hugging Face Transformers; not deployed by a Hugging Face Inference Provider
View model
→
