Alibaba-backed Moonshot AI has officially launched the Kimi K2.5 AI model, marking a major step forward in multimodal and agentic artificial intelligence. Announced on January 27, the open-source release introduces two standout capabilities: vision-based coding and a powerful self-directed agent swarm system.

The company says these features are designed to boost productivity for both everyday users and enterprise knowledge workers—while keeping compute requirements far lower than competing frontier models.

A Unified Multimodal Architecture

At the core of the Kimi K2.5 AI model is a single multimodal architecture that combines text and visual inputs. That allows the system to handle reasoning, coding, visual understanding, and autonomous task execution without switching between tools or models.

Kimi K2.5 is available through the Kimi app, Kimi.com, APIs, and Kimi Code, with both Instant and Thinking modes supported.

“We aim to turn your ideas into products and your data into insights, while minimising technical friction,” said Moonshot AI founder and CEO Zhilin Yang during the launch presentation.

Agent Swarms That Work in Parallel

One of the most notable upgrades in the Kimi K2.5 AI model is its agent swarm capability. The system can autonomously create and coordinate up to 100 sub-agents, running in parallel across as many as 1,500 steps.

Unlike traditional workflows that require predefined roles or rigid task plans, Kimi’s Swarm Mode dynamically adapts as tasks evolve—allowing complex, multi-step problems to be completed faster. Swarm Mode is currently available on Kimi.com for Allegretto-tier users and above.

Moonshot AI says this approach significantly improves real-world task completion, especially for research, analysis, and enterprise workflows.

Vision-Based Coding With “An Eye for Design”

Kimi K2.5 is also Moonshot AI’s most capable open-source system yet for coding with vision. By reasoning directly over images and videos, the model supports image-to-code replication, visual debugging, and front-end development using screenshots or screen recordings.

That means users can turn visual designs into functional websites or applications—simply by showing Kimi what they want built.

“We didn’t just want Kimi to code; we wanted it to have an eye for design,” Yang said.

Through Kimi Code, the model supports full-stack development and integrates with terminals and popular IDEs such as VS Code, Cursor, and Zed, while accepting multimodal inputs.

Strong Benchmarks, Lower Compute

Moonshot AI claims the Kimi K2.5 AI model sets new benchmark results across agentic, coding, and vision tasks, including Humanity’s Last Exam, BrowseComp, and DeepSearchQA. According to the company, it delivers performance comparable to leading AI systems—while using significantly less compute.

That efficiency is no accident. Moonshot AI president Yutong Zhang said the company’s limited access to large-scale compute forced a sharper focus on fundamentals.

“We’re developing frontier models with roughly 1% of the resources used by major U.S. labs,” Zhang said at the World Economic Forum in Davos last week.

Rapid Growth and Enterprise Focus

The launch builds on the momentum of the Kimi K2 series, which emphasizes reasoning and agentic intelligence. Moonshot AI reports that global paid users grew at an average month-on-month rate of more than 170% between September and November 2025, while API revenue quadrupled following the release of Kimi K2 Thinking.

The Kimi K2.5 AI model is designed to automate common knowledge-worker tasks such as financial modelling, document formatting, and presentation creation. Moonshot AI says tasks like merging large reports or converting long research papers into slide decks can now be completed in minutes through conversational interaction.

Expanding Ecosystem and What’s Next

Kimi K2.5 is also available via third-party platforms, including NVIDIA’s build.nvidia.com and Fireworks. Moonshot AI says it will continue investing heavily in agentic intelligence, with a focus on reliable real-world execution under time and resource constraints.

As competition in open-source AI accelerates, the Kimi K2.5 AI model positions Moonshot AI as a serious contender—especially for users who want powerful agentic systems without the massive compute bill.

Don’t miss out on our latest news—follow us for the latest AI news, breakthroughs, and insights that matter.