WordPress.com gives AI agents the power to create edit and manage your site content
March 21, 2026
CNN starts building its own AI agents to handle future media deals
April 6, 2026
Cursor rolled out Composer 2 this week with big claims of frontier level coding intelligence that could handle complex software tasks better than rivals. The AI coding company positioned the model as a major step forward for its popular editor yet left out one key detail in the announcement. It built Composer 2 on top of Moonshot AI‘s recently released open source Kimi K2.5 model.
An X user named Fynn spotted the connection first by inspecting API responses and noting model identifiers that pointed straight to Kimi. The post quickly gained traction and forced a response from Cursor executives. Lee Robinson, vice president of developer education at Cursor, replied that yes the new model started from an open source base but only about one quarter of the total compute went to that foundation while the rest came from Cursor‘s own continued training. He added that benchmark results for Composer 2 look very different from the original Kimi scores as a result. 🤖
What the partnership actually means 🔗
Moonshot AI itself confirmed the arrangement on X and described it as an authorized commercial partnership routed through Fireworks AI. The company expressed pride that its Kimi K2.5 served as the starting point and praised Cursor for applying additional pretraining and high compute reinforcement learning. This setup fits the broader open model ecosystem where developers can legally build on top of released weights as long as license terms are followed. For context on Kimi K2.5 capabilities see coverage from InfoQ on Moonshot AI’s open weight release.
Cursor co founder Aman Sanger later called the omission a miss and said the team would mention the base model more clearly in future announcements. The startup sits in a strong market position after raising 2.3 billion dollars last fall at a 29.3 billion dollar valuation and reportedly surpassing 2 billion dollars in annualized revenue. Still the episode shows how fast scrutiny arrives in public technical discussions.
Why the base model choice matters in today’s AI race 🌍
Building on an open source foundation from Moonshot AI a Chinese company backed by Alibaba and HongShan adds another layer to conversations about global AI development. Cursor emphasized that its final product delivers distinct performance yet the initial reliance on Kimi K2.5 underscores how quickly strong base models travel across borders and companies. Developers who rely on Cursor for daily work now have clearer insight into the stack powering their agentic coding sessions.
The story also ties into larger trends around specialized coding tools. Readers interested in comparing options can explore the top AI coding and development tools roundup or the dedicated Cursor profile page for feature breakdowns. Additional context on the underlying model appears in the Kimi assistant overview.
Geopolitical sensitivities around AI progress add weight to transparency questions but the practical outcome remains the same. Cursor users gain access to a capable model while the open ecosystem continues to reward rapid iteration on strong foundations. The episode serves as a reminder that in fast moving AI coding the full picture often emerges after launch rather than in the press release.
