Coding Platform Cursor Admits Use of China’s Kimi K2.5 Model in Composer 2 After Backlash

Popular AI code editor, Cursor, has confirmed that its newly launched Composer 2 coding model is built on top of Moonshot AI’s Kimi K2.5, addressing a wave of developer speculation and criticism that followed the model’s debut earlier this week.

The disclosure came after users on platforms like X and Reddit began examining Cursor’s API calls and model identifiers, uncovering references to Kimi variants such as “kimi-k2p5-rl-0317-s515-fast.” 

Initially, founders at Moonshot AI publicly raised concerns on X about Cursor potentially not complying with licensing terms or fees related to its use of Kimi K2.5 in Composer 2, before later deleting those posts.

was messing with the OpenAI base URL in Cursor and caught this

accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast

so composer 2 is just Kimi K2.5 with RL

at least rename the model ID https://t.co/MQOuEuF3Pd pic.twitter.com/fyUWbo1InF

— Fynn (@fynnso) March 19, 2026

Cursor Co-founder Aman Sanger, in a post on X, acknowledged the omission, saying the company had tested multiple base models and found Kimi K2.5 to be the strongest on perplexity-based evaluations. “The combination of the strong base, CPT and RL, and Fireworks’ inference and RL samplers make Composer-2 frontier level. It was a miss to not mention the Kimi base in our blog from the start. We’ll fix that for the next model,” he wrote.

Cursor accesses the Kimi K2.5 model through Fireworks AI, which provides hosted inference and reinforcement learning infrastructure as part of a commercial partnership. 

Moonshot AI, the Chinese company behind Kimi, publicly endorsed the launch, calling it a strong example of how open models can be extended through downstream training.

“Congrats to the Cursor team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support,” it wrote on X.

Composer 2, announced on March 19, is positioned as Cursor’s most advanced in-house coding model to date. It features a context window of around 200,000 tokens and is built for agentic workflows, including multi-file edits, terminal operations, and tool use.

Cursor claims the model delivers “frontier-level” coding performance, citing gains on benchmarks such as Terminal-Bench and SWE-Bench Multilingual, where it reportedly outperforms competitors like Claude Opus 4.6 in select scenarios while remaining more cost-efficient.