Moonshot AI

Moonshot AI announced Kimi K2.5 on its API platform, describing it as its most capable engine yet for sophisticated agents, visual reasoning, and professional-grade coding tools. The release also simplified the lineup by making Turbo-level performance the default.

• Most capable Kimi engine at launch
• Built for agents and tool-calling
• Native vision and visual reasoning
• Strong frontend and coding performance
• Turbo-level performance became the default

Moonshot AI officially released Kimi K2, positioning it as a high-speed model with strong reasoning and tool-use capabilities. The launch formed the base of the later Kimi K2 turbo and Kimi K2.5 ecosystem.

• Official Kimi K2 model release
• Base for later Kimi turbo variants
• Reasoning and tool-use focus
• Important platform milestone for Moonshot AI

Moonshot AI released Kimi k1.5, demonstrating that reinforcement learning with long chain-of-thought reasoning can achieve state-of-the-art performance on complex tasks. The model showed strong reasoning capabilities through extensive RL training, achieving competitive results against top models on math, coding, and reasoning benchmarks without relying on massive supervised datasets.

• Long chain-of-thought reasoning
• Extensive reinforcement learning
• Strong math and coding performance
• Competitive with top models
• Minimal supervised data needed