Moon's Dark Side Open Platform Launches New Model kimi-latest to Meet Diversified Needs

Dark Moon Corporation has officially released its latest visual model, Kimi-Latest. This model is part of the Kimi intelligent assistant product line and leverages cutting-edge Kimi large model technology, although it may include some features that are still in development.

Kimi-Latest demonstrates uniqueness in several areas. First, its context processing capability has been significantly improved, supporting up to 128k tokens of context length. It automatically selects between 8k, 32k, or 128k models based on user needs for billing purposes. Additionally, as a visual model, Kimi-Latest supports image understanding, offering users more interactive options.

In terms of cost, Kimi-Latest introduces an automatic context caching mechanism to reduce expenses. When the cache is hit, token costs are only 1 yuan RMB per megabyte. However, manual context caching is not currently supported.

Besides these features, Kimi-Latest retains some functionalities from the moonshot-v1 series, including ToolCalls, JSON Mode, Partial Mode, and internet search capabilities. This allows users to enjoy the benefits of the new model while maintaining familiarity with existing features.

In terms of application scenarios, Kimi-Latest is ideal for users who are utilizing large model chat applications and seek an experience similar to that of the Kimi intelligent assistant. It is also a viable option for businesses developing AI assistants or customer service systems.

However, if a user's tasks involve intent recognition or structured data extraction, the moonshot-v1 series might be more suitable. For users already satisfied with moonshot-v1 series models and stable prompt performance, continuing with this series could be more prudent, as switching to Kimi-Latest may require revising prompts.

Additionally, Dark Moon Corporation notes that Kimi-Latest currently only supports calling the standard Kimi model. The Kimi k1.5 long-thinking model does not support API calls. Users interested in research or evaluation can submit a test application via a form. The company will assess applications and provide API keys via email for accessing the Kimi k1.5 model upon approval.