Archive — US — techcrunch
China’s Moonshot releases a new open-source model Kimi K2.5 and a coding agent
China’s Moonshot AI has launched a new open-source model, Kimi K2.5, which excels in text, image, and video understanding, outperforming competitors in coding benchmarks. The company also unveiled Kimi Code, an open-source coding tool for developers, aiming to rival major players in the AI coding sector. Moonshot recently secured funding, raising its valuation significantly amidst a competitive AI landscape.
Key points
- Kimi K2.5 is an open-source model trained on 15 trillion visual and text tokens.
- The model performs well in coding tasks and agent orchestration.
- Benchmark tests reveal Kimi K2.5 outperforms Gemini 3 Pro and GPT 5.2 in various tasks.
- Kimi Code, a new open-source coding tool, allows input from images and videos.
- Moonshot AI has received $1 billion in Series B funding and is seeking more at a higher valuation.
Why it matters
- AI coding tools are growing rapidly, predicted to contribute significantly to revenue by 2025.
- The launch of Kimi K2.5 aligns with trends of increased multimodal AI capabilities.
- Competitors like Deepseek are releasing similar coding models, fostering intense competition in AI.
Moonshot's Kimi K2.5 and Kimi Code represent strategic advancements in AI, particularly in coding and multimodal applications, positioning the company favorably in an evolving tech landscape.