Moonshot AI Preps Kimi K2 Model With 256K Context

Moonshot AI, a Beijing-based startup backed by Alibaba and Tencent, is preparing to launch an updated version of its Kimi K2 AI model.

The company briefly announced the Kimi K2-0905 beta test on Discord before editing the post to remove technical details. According to growth team member Aspen Choong, the update will offer improved coding and creative writing skills, a larger 256,000-token context window, and fewer hallucinations. However, planned reasoning and vision features will not be included, and beta testing has been delayed due to API issues. Founded in March 2023, Moonshot AI has already achieved a $3.3 billion valuation, attracting major funding from Alibaba and Tencent.

Key Highlights

• Moonshot AI valued at $3.3B, backed by Alibaba and Tencent.
• New Kimi K2 model expands to 256K tokens with fewer hallucinations.
• Beta delayed due to API issues, rollout limited to 20 users

Its original Kimi K2 model ranks tied for eighth overall and fourth in coding on developer leaderboard LMArena.

The expansion from 128,000 to 256,000 tokens marks progress, but Chinese AI startups still lag behind global leaders. Moonshot’s founder Yang Zhilin acknowledged the gap, noting that “256K context is simply not enough,” as competitors like Google DeepMind’s Gemini 2.5 Pro already support 1 million tokens. Larger context windows are crucial for complex tasks like document analysis and codebase management.

Also Read: Alibaba Launches Qwen3-Coder, AI Model for Code Generation

The beta’s Discord-based rollout, limited to 20 users, shows caution amid concerns that API-related issues may undermine deployment reliability. For AI companies, robust serving infrastructure is increasingly essential to ensure scalability and maintain market competitiveness in the fast-moving AI development race.