Model Overview
Kimi k1.5 is a multimodal reasoning model from Moonshot AI, trained using reinforcement learning. The official research report discusses long and short chain-of-thought methods, as well as joint training on text and visual data. (github.com)
Capabilities and Uses
The model is designed for tasks such as mathematical reasoning, programming, and visual understanding, and can be used for problem-solving, code analysis, and reasoning over images and text. The benchmark results presented in the official report apply to specific evaluation conditions and do not guarantee the same performance across all real-world tasks. (github.com)
Limitations
Outputs may be incorrect; users should verify important conclusions and code. The 128K mentioned by the official source is the context length used in reinforcement learning training and cannot be taken as the available context window of this preview API. The API's current availability and specific interface limitations have not been verified. (github.com)
Dates and Source Notes
- Model announcement date: 2025-01-20. An official GitHub commit introduced Kimi k1.5 on that date; this is the commit date of the model announcement, not a verified initial release or listing date for the
kimi-k1.5-previewAPI. (github.com) - Source note: In the current data, the specific initial release date and month of
kimi-k1.5-previewhave not been independently verified. The official announcement did not list this API ID; a research appendix listed the ID as a Moonshot AI model and recorded its data collection date as February 2025, but the collection date cannot be assumed to be the initial release date. Therefore, no initial release date or end-of-month placeholder date is provided for this preview version. (github.com)
Pricing
As of 2026-11-20, the official input and output pricing in RMB or USD for this specific preview version has not been verified; prices for other models are not used as substitutes. The official domestic pricing page and the official international pricing page have been consulted. (platform.kimi.com)