project
Kimi Latest - Kimi's real-time updated AI model, synchronized with the Kimi smart assistant.
Kimi Latest is a real-time updated AI model launched by Kimi, providing users with a model experience that is synchronized with the Kimi intelligent assistant. It supports a 128k context length and can automatically select 8k, 32k, or 128k models based on input...
What is Kimi Latest?
Kimi Latest is a real-time updated AI model launched by Kimi, providing users with a model experience that is synchronized with the Kimi intelligent assistant. It supports a 128k context length and can automatically select 8k, 32k, or 128k models for billing based on input. It also has image understanding capabilities and can handle visual tasks. Kimi Latest supports automatic context caching, with extremely low token fees for cache hits, only 1/M token.
Kimi Latest's main functions
- Latest model capabilitiesIt is always benchmarked against the latest model currently used by the Kimi Smart Assistant, including experimental features that are not yet stable.
- Long context supportSupports 128k context length and can automatically select 8k, 32k or 128k models for billing based on input.
- Image comprehension abilityAs a visual model, it supports image understanding capabilities.
- Automatic context cachingSupports automatic context caching, with a cost of only 1/M tokens for cached tokens.
- Advanced feature integrationSupports ToolCalls, JSON Mode, Partial Mode, and online search functionality.
- Wide range of applicable scenariosSuitable for large-scale chat applications (such as ChatWise, ChatBox, etc.), providing a chat experience similar to the Kimi smart assistant.
Kimi Latest's technical principles
- Long-CoT (Long-CoT) technologyThe Kimi Latest model inherits the long thought chain technology from Kimi K1.5. By extending the length of the reasoning chain, the model can perform deeper thinking and more complex reasoning. Compared to traditional short reasoning chains, long thought chains significantly improve the model's performance in mathematical, programming, and multimodal reasoning tasks.
- Long Context ScalingKimi Latest supports context lengths up to 128k and optimizes computational efficiency through a partial rollout strategy. The model is more advantageous when handling long texts and complex dialogues.
- Improved Policy OptimizationThe model employs a reinforcement learning formula based on long thought chains, combined with online mirror descent for policy optimization.
- Multimodal joint trainingKimi Latest supports multimodal information processing, including joint training with text and visual data. This enables the model to process both text and visual information simultaneously, achieving cross-modal inference.
- Automatic context cachingKimi Latest introduces automatic context caching technology, where the cost of cached tokens is only 1/M tokens.
- Simple framework designThe model abandons complex techniques such as Monte Carlo tree search and value function, and achieves powerful inference capabilities by expanding the context length and optimization strategies.
- Long2Short technologyKimi Latest uses Long2Short technology to transfer the reasoning capabilities of long-chain thinking models to short-chain thinking models, further optimizing performance.
Kimi Latest's project address
- Project official websiteKimi Smart Assistant
Kimi Latest's model pricing
- kimi-latest-8k: 1M tokens; Input price (cache hit) ¥1.00; Input price (cache miss) ¥12.00; Output price ¥12.00.
- kimi-latest-32k: 1M tokens; Input price (cache hit) ¥1.00; Input price (cache miss) ¥24.00; Output price ¥24.00.
- kimi-latest-128k: 1M tokens; Input price (cache hit) ¥1.00; Input price (cache miss) ¥60.00; Output price ¥60.00.
Why launch Kimi Latest?
- Meeting differentiated needsThe needs of users of the Kimi intelligent assistant and the open platform are gradually diverging. The intelligent assistant requires responses with greater emotional value, while open platform users prioritize the model's structured text processing and accurate output. This difference may lead to issues such as ineffective prompts during use.
- Maintain model stabilityKimi Assistant's rapid iteration feature constantly introduces new functions, but it may also introduce some experimental features that are not yet stable. While these features may perform well in some scenarios, they may have problems in others. Therefore, the model on the open platform differs from the model used by the assistant, raising questions from users about inconsistent responses across different platforms.
- Synchronize with the latest capabilitiesTo address the aforementioned issues, the Kimi Latest model was introduced. It will always be aligned with the latest model currently used by the Kimi Smart Assistant, and will be upgraded synchronously with the Smart Assistant's updates. This way, users of the open platform can experience the superior performance of the latest model while maintaining the stability of the existing model.
- Improve user experienceBy launching the Kimi Latest model, Moonlight Technology provides developers and users with a more flexible, efficient, and stable AI model solution to meet their diverse needs in different scenarios.
Special Notes on Kimi Latest
- If you are using large-scale chat applications such as ChatWise, ChatBox, LobeChat, and OpenCat, and want to use Kimi as your AI chat partner, the Kimi Latest model is the best choice.
- To build an AI-powered intelligent assistant or customer service for your product, and to achieve a chat experience similar to Kimi's intelligent assistant, choose the Kimi Latest model.
- For large models used for intent recognition or structured data extraction, please select the moonshot-v1 series models.
- The effect is stable when using the moonshot-v1 series model, so you can continue to use it. If you switch to the Kimi Latest model, you need to readjust the prompt words.
Application scenarios of Kimi Latest
- Language learningIt helps users practice language skills by providing translation, grammar correction, and writing guidance.
- Teaching aidsIt helps teachers design course outlines, create teaching materials, and assess student assignments.
- Market ResearchAnalyze market trends and competitor information, and provide market research reports.
- Content creationGenerate copy, advertising slogans, news reports, etc., to improve content creation efficiency.
- Image recognition and analysisIt identifies objects, scenes, and text in images and provides image descriptions and analysis reports.