China's Moonshot Unveils New Open-Source Model Kimi K2.5 and AI Coding Agent
Moonshot AI, a Chinese company backed by Alibaba and HongShan (formerly Sequoia China), has introduced Kimi K2.5, a new open-source model capable of processing text, images, and video.
The company revealed the model was trained on 15 trillion mixed visual and text tokens, enabling native multimodality. It excels at coding tasks and managing agent swarms
—a setup where multiple AI agents collaborate. In benchmark tests, Kimi K2.5 matches the performance of leading proprietary models and even surpasses them in certain areas.
For example, on the SWE-Bench Verified coding benchmark, Kimi K2.5 outperforms Gemini 3 Pro. It also scores higher than GPT 5.2 and Gemini 3 Pro on the SWE-Bench Multilingual test. In video comprehension, it beats GPT 5.2 and Claude Opus 4.5 on VideoMMMU (Video Massive Multi-discipline Multimodal Understanding), a benchmark evaluating a model's reasoning abilities with video content.

Image Credits: Moonshot AI Moonshot AI noted that while the model handles text effectively, users can also input images or videos and request it to create a similar interface based on that media.
Here's a short video from our founder, Zhilin Yang.
(It's his first time speaking on camera like this, and he really wanted to share Kimi K2.5 with you!) pic.twitter.com/2uDSOjCjly
— Kimi.ai (@Kimi_Moonshot) January 27, 2026
To make these coding features accessible, the company launched Kimi Code, an open-source coding tool that competes with Anthropic’s Claude Code and Google’s Gemini CLI. Developers can use Kimi Code via their terminal or integrate it with IDEs like VSCode, Cursor, and Zed. The startup also confirmed that Kimi Code accepts images and videos as input.
AI coding tools are rapidly gaining traction and becoming significant revenue sources for AI companies. Anthropic announced in November that Claude Code had reached $1 billion in annualized recurring revenue (ARR). A Wired report earlier this month indicated the tool added another $100 million to that total by the end of 2025. Meanwhile, Moonshot's Chinese rival, Deepseek, is preparing to launch a new model with advanced coding capabilities next month, according to The Information.
Moonshot was founded by Yang Zhilin, a former AI researcher at Google and Meta. The company secured $1 billion in a Series B funding round, achieving a $2.5 billion valuation. Bloomberg reported it recently raised an additional $500 million at a $4.3 billion valuation and is already seeking a new funding round that would value it at $5 billion.
Related article
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
AI race won’t wait, nor will this TechCrunch Disrupt 25% discount
Every founder building in AI right now is asking the same question: Are we moving fast enough, and effectively enough? TechCrunch Disrupt 2026 puts you in the same room as the people wrestling with that existential question at the highest levels. Fro
Related Special Topic Recommendations
Comments (1)
0/500
Klingt spannend! Ich finde vor allem den Multimodal-Ansatz (Text/Bild/Video) von Kimi K2.5 interessant – das sind genau die Fähigkeiten, die KI für reale Aufgaben braucht. Obwohl es viele Open-Source-Modelle gibt, heben sich 15 Billionen Tokens an Trainingsdaten schon hervor. Vielleicht sollte ich es mal mit meinem Codierprojekt testen... 👨💻
Moonshot AI, a Chinese company backed by Alibaba and HongShan (formerly Sequoia China), has introduced Kimi K2.5, a new open-source model capable of processing text, images, and video.
The company revealed the model was trained on 15 trillion mixed visual and text tokens, enabling native multimodality. It excels at coding tasks and managing agent swarms
—a setup where multiple AI agents collaborate. In benchmark tests, Kimi K2.5 matches the performance of leading proprietary models and even surpasses them in certain areas.
For example, on the SWE-Bench Verified coding benchmark, Kimi K2.5 outperforms Gemini 3 Pro. It also scores higher than GPT 5.2 and Gemini 3 Pro on the SWE-Bench Multilingual test. In video comprehension, it beats GPT 5.2 and Claude Opus 4.5 on VideoMMMU (Video Massive Multi-discipline Multimodal Understanding), a benchmark evaluating a model's reasoning abilities with video content.

Moonshot AI noted that while the model handles text effectively, users can also input images or videos and request it to create a similar interface based on that media.
Here's a short video from our founder, Zhilin Yang.
— Kimi.ai (@Kimi_Moonshot) January 27, 2026
(It's his first time speaking on camera like this, and he really wanted to share Kimi K2.5 with you!) pic.twitter.com/2uDSOjCjly
To make these coding features accessible, the company launched Kimi Code, an open-source coding tool that competes with Anthropic’s Claude Code and Google’s Gemini CLI. Developers can use Kimi Code via their terminal or integrate it with IDEs like VSCode, Cursor, and Zed. The startup also confirmed that Kimi Code accepts images and videos as input.
AI coding tools are rapidly gaining traction and becoming significant revenue sources for AI companies. Anthropic announced in November that Claude Code had reached $1 billion in annualized recurring revenue (ARR). A Wired report earlier this month indicated the tool added another $100 million to that total by the end of 2025. Meanwhile, Moonshot's Chinese rival, Deepseek, is preparing to launch a new model with advanced coding capabilities next month, according to The Information.
Moonshot was founded by Yang Zhilin, a former AI researcher at Google and Meta. The company secured $1 billion in a Series B funding round, achieving a $2.5 billion valuation. Bloomberg reported it recently raised an additional $500 million at a $4.3 billion valuation and is already seeking a new funding round that would value it at $5 billion.
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
AI race won’t wait, nor will this TechCrunch Disrupt 25% discount
Every founder building in AI right now is asking the same question: Are we moving fast enough, and effectively enough? TechCrunch Disrupt 2026 puts you in the same room as the people wrestling with that existential question at the highest levels. Fro
Klingt spannend! Ich finde vor allem den Multimodal-Ansatz (Text/Bild/Video) von Kimi K2.5 interessant – das sind genau die Fähigkeiten, die KI für reale Aufgaben braucht. Obwohl es viele Open-Source-Modelle gibt, heben sich 15 Billionen Tokens an Trainingsdaten schon hervor. Vielleicht sollte ich es mal mit meinem Codierprojekt testen... 👨💻





Home






