Cursor Confirms Flagship Composer Model Uses Moonshot’s Open-Source Kimi K2.5
Cursor confirmed that its flagship coding model, Composer 2 and 2.5, is built on Moonshot AI’s open-source Kimi K2.5 base model after a developer spotted a Kimi model ID in the company’s API responses. Cursor said it started from the Kimi base using about 25% of the compute, then added reinforcement learning and its own data and training on top.
The disclosure indicates that Cursor is relying on an open-source foundation and then tailoring it with product-specific reinforcement learning, training data and agent features. It also suggests the company is using open bases such as Kimi, Qwen and DeepSeek to build cheaper, faster specialized variants and reduce reliance on closed models from Anthropic and OpenAI.
From the sources (2 posts)
@brianroemmele@DefNotZap Cursor’s flagship coding model (Composer 2 / 2.5) is literally built on the open-source Kimi K2.5 from Moonshot AI. A developer quickly spotted this via the model ID in Cursor’s API responses (kimi-k2p5-rl-...), and Cursor’s tea
@brianroemmeleRT @BrianRoemmele: @DefNotZap Cursor’s flagship coding model (Composer 2 / 2.5) is literally built on the open-source Kimi K2.5 from Moonsh…