Kimi-K2.5 is Moonshot AI’s most versatile model to date, featuring a native multimodal architecture that simultaneously supports visual and text input, thinking and non-thinking modes, and conversational and Agent tasks. With its 256K ultra-long context window, multimodal understanding, and Tool Calling capabilities, it sets a new benchmark in open-source visual programming and Agent clusters, aiming to empower developers to build next-generation AI applications.
- Native Multimodal Architecture: Supports mixed input of visual and text, capable of image recognition, visual programming, and other tasks, leading the way in open-source visual programming.
- 256K Ultra-Long Context Window: Provides a 256,000 token context window, supporting long-form reasoning and complex task processing, effectively handling large amounts of information.
- Agent Clusters and Tool Calling: Supports a preview version of Agent clusters, capable of supporting up to 100 sub-agents and 1,500 tool calls, operating 4.5 times faster than single-agent configurations, significantly enhancing automation and complex task execution efficiency.
- Exceptional Coding Capabilities: Performs outstandingly in benchmarks such as SWE-Bench and LiveCodeBench, leading in competitive programming, and is priced significantly lower than comparable models.
- Thinking Modes: Offers both thinking and non-thinking modes, allowing the model to engage in deeper reasoning and planning when needed.
- Visual Programming and Automation: From pixel-level webpage replication to expert-level office delivery, efficiently handles complex tasks, especially suitable for scenarios requiring visual understanding and programming capabilities.
- Ultra-Long Text Processing and Analysis: Applicable to tasks requiring the processing of massive amounts of information, such as legal document review, research report analysis, and codebase understanding.
- Agent Tasks and Multi-Agent Collaboration: Builds complex automated workflows, enabling multi-agent collaboration to complete large-scale tasks.
- Professional Code Generation and Debugging: Provides powerful code generation, optimization, and debugging capabilities for developers.
| Capability | Detailed Description |
|---|
| Reasoning Ability | Extremely Strong. Excels in long-context reasoning, Agent task planning, and execution. |
| Creative Ability | Extremely Strong. Particularly adept at visual programming, code generation, and multimodal content understanding. |
| Multimodal Ability | Native multimodal, supports visual and text input, outstanding in visual programming. |
| Response Speed | Fast response in quick mode, efficient parallel processing achievable in Agent cluster mode. |
| Context Window | 256,000 Tokens |
| Max Output | 256,000 Tokens |
| Model | Input (Credits/Token) | Output (Credits/Token) |
|---|
| Kimi-K2.5 | 0.23 | 3.00 |
