Moonshot AI's Kimi K2.5 Joins Microsoft Foundry with Enhanced Multimodal Capabilities

Moonshot AI's Kimi K2.5, the company's next-generation multimodal and agentic model, has been integrated into Microsoft Foundry, expanding the platform's AI model offerings with advanced vision-language capabilities and improved execution efficiency.

The latest release represents a significant advancement in multimodal AI technology, with Kimi K2.5 delivering enhanced performance across several key areas. The model has been pre-trained with an additional 15 trillion vision-text tokens, enabling stronger image and video understanding capabilities, improved optical character recognition (OCR), and more sophisticated multimodal question answering.

One of the standout features of Kimi K2.5 is its Agent Swarm execution capability, which can orchestrate up to 100 parallel agents and handle 1,500 tool calls simultaneously. This parallel processing approach reduces execution time by up to 4.5 times compared to sequential workflows used in previous versions, making it particularly valuable for complex, multi-step tasks that require coordination between multiple AI agents.

The model's image and video to code capabilities have also seen substantial improvements. Developers can now leverage Kimi K2.5 for visual debugging and reconstructing user interfaces from screenshots or video snippets, streamlining the development process for applications that need to interpret visual inputs and generate corresponding code.

According to Moonshot AI's benchmarks, Kimi K2.5 achieves state-of-the-art results across multiple domains:

• AIME 2025: 96.1% accuracy in mathematical problem-solving
• MMLUPro: 87.1% performance in general knowledge tasks
• MMMUPro (Vision): 78.5% accuracy in multimodal reasoning

These results position Kimi K2.5 as a competitive option in the growing landscape of advanced AI models, particularly for applications requiring sophisticated multimodal understanding and execution capabilities.

Microsoft Foundry users can access Kimi K2.5 with the following pricing structure:

• Input tokens: $0.60 per 1 million tokens
• Output tokens: $3.00 per 1 million tokens

This pricing places Kimi K2.5 in the premium segment of AI models available through Microsoft Foundry, reflecting its advanced capabilities and performance characteristics.

The integration of Kimi K2.5 into Microsoft Foundry provides developers with another powerful tool in their AI toolkit, particularly for applications that require sophisticated vision-language integration, complex agent orchestration, or advanced multimodal coding workflows. As businesses continue to explore AI applications that bridge visual and textual understanding, models like Kimi K2.5 offer compelling capabilities for next-generation AI-powered solutions.