Moonshot AI releases Kimi K2.5 with Agent Swarm that coordinates 100 parallel AI agents

China's Moonshot releases a new open-source model Kimi K2.5 and a coding agent

China's Moonshot AI, which is backed by the likes of Alibaba and HongShan (formerly Sequioa China), today released a new open-source model, Kimi K2.5, which understands text, image, and video. The company said that the model was trained on 15 trillion mixed visual and text tokens, and that's why it is natively multimodal. It added that the models are good at coding tasks and handling agent swarms -- an orchestration where multiple agents work together. In released benchmarks, the model matches the performance of the proprietary peers and even beats them in certain tasks. For instance, in the coding benchmark, the Kimi K2.5 outperforms Gemini 3 Pro at the SWE-Bench Verified benchmark, and scores higher than GPT 5.2 and Gemini 3 Pro on the SWE-Bench Multilingual benchmark. In video understanding, it beats GPT 5.2 and Claude Opus 4.5 on VideoMMMU (Video Massive Multi-discipline Multimodal Understanding), a benchmark that measures how a model reasons over videos. Moonshot AI said that on the coding front, while the model can understand text well, users can also feed it images or videos and ask it to make a similar interface shown in those media files. To let people use these coding capabilities, the company has launched an open-source coding tool called Kimi Code, which would rival Anthropic's Claude Code or Google's Gemini CLI. Developers can use Kimi Code through their terminals or integrate it with development software such as VSCode, Cursor, and Zed. The startup said that developers can use images and videos as input with Kimi Code. Coding tools have gained rapid popularity and are becoming revenue drivers for AI labs. Anthropic announced in November that Claude Code had reached $1 billion in annualized recurring revenue (ARR). Earlier this month, Wired reported that by the end of 2025, the tool had added $100 million to that figure. Moonshot's Chinese competitor, Deepseek, is set to release a new model with strong coding chops next month, according to a report by The Information. Moonshot was founded by former Google and Meta AI researcher Yang Zhilin. The company raised $1 billion in funding in a Series B round at a $2.5 billion valuation. According to Bloomberg, the startup picked up $500 million in funding last month at $4.3 billion valuation. What's more, the report noted that it is already seeking to raise a new round at a $5 billion valuation.

Moonshot's new Kimi K2.5 model can build websites from visual inputs - here's how it works

Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways * Moonshot debuted its open-source Kimi K2.5 model on Tuesday. * It can generate web interfaces based solely on images or video. * It also comes with an "agent swarm" beta feature. Alibaba-backed Chinese AI startup Moonshot released Kimi K2.5 on Tuesday, describing it in a blog post as the world's "most powerful open-source model to date." Built on top of the Kimi K2 LLM, which debuted last summer, Moonshot's latest model comes with coding capabilities that could make it a serious competitor with its proprietary counterparts. Kimi K2.5 scored comparably to frontier models from OpenAI, Google, and Anthropic on the SWE-Bench Verified and SWE-Bench Multilingual coding benchmarks, according to data published by Moonshot. Its ability to create front-end web interfaces from visual inputs, however, is what could truly set it apart from the crowd. Coding with vision Kimi K2.5 was pretrained with 15 trillion text and visual tokens, making it "a native multimodal model," according to Moonshot, that can generate web interfaces from uploaded images or video, complete with interactive elements and scroll effects. In a demo video of this "coding with vision" capability included in Moonshot's blog post, Kimi K2.5 generated a draft of a new website based on a recorded video of a preexisting website, shown from the perspective of a user's screen as they scroll. The model was able to recreate the general aesthetic, even if -- in classic AI style -- it made some slight visual blunders along the way, like depicting continents on a globe as amorphous blobs. It's unclear how practical this kind of capability will be. (Why would a company need to create a slightly less visually appealing AI-generated copy of an already perfectly reasonable website?) Still, generating mock-ups of websites and apps exclusively from images or videos would mark a meaningful step forward for so-called "vibe coding" tools, which are based on intuitive methods easily deployed by non-experts rather than traditional coding. ChatGPT, Claude, and Gemini can generate raw code for new web assets based on screenshots or other images, but that still leaves the user needing to translate it into a finished and usable product. The novelty (and potential market value) of Moonshot's new model is that it cuts out that intermediary step. "By reasoning over images and video, K2.5 improves image/video-to-code generation and visual debugging, lowering the barrier for users to express intent visually," the company wrote in its blog post. Also: I used Claude Code to vibe code a Mac app in 8 hours, but it was more work than magic If it proves useful in the real world, especially among businesses, other developers will probably follow suit with similar capabilities for their own models. Kimi K2.5's coding capabilities have been made available through an open source platform called Kimi Code, which can be accessed through integrated development environments (IDEs) like Cursor, VSCode, and Zed. The new model is also available through Kimi.com, the Kimi App, and the Kimi API. Agent swarm Moonshot also unveiled a research preview called "agent swarm," which orchestrates up to one hundred "sub-agents" to improve performance on certain multistep tasks. By running multiple tasks in parallel to one another, agent swarm can also speed up the compute process. "Running these subtasks concurrently significantly reduces end-to-end latency compared to sequential agent execution," Moonshot wrote in its blog post, adding that internal evaluations showed that end-to-end runtime -- the total process from input to the completion of the final output -- could be reduced by up to 80%. Also: I used Claude Code to vibe code an Apple Watch app in just 12 hours - instead of 2 months Users with an active "Allegretto" or "Vivace" Moonshot account (costing $31/month and $159/month, respectively) can give agent swarm a try on the Kimi website by clicking the model drop-down menu on the bottom-right of the prompt box and selecting "K2.5 Agent Swarm (Beta)."

Moonshot AI debuts Kimi K2.5, most powerful open source LLM beating Opus 4.5 with swarm of parallel agents

Chinese company Moonshot AI upgraded its open-sourced Kimi K2 model, transforming it into a coding and vision model with an architecture that supports an Agent Swarm orchestration. The new model, Moonshot Kimi K2.5, could offer a compelling orchestration option for enterprises that, instead of building an overarching framework that decides when agents complete a task, want agents to pass off actions to the next agent automatically. The company characterized Kimi K2.5 as an "all-in-one model" that supports both visual and text inputs, allowing users to leverage the model for more "visual" coding projects. Moonshot did not publicly disclose K2.5's parameter count, but the Kimi K2 model, which it is based on, had 1 trillion total parameters and 32 billion activated parameters thanks to its mixture-of-experts architecture. This is the latest open-source model to offer an alternative to the more closed options from Google, OpenAI and Anthropic, and it outperforms them on key metrics including agentic workflows, coding, and vision. On the Humanity's Last Exam (HLE) benchmark, Kimi K2.5 scored 50.2% (with tools), surpassing OpenAI's GPT-5.2 (xhigh) and Claude Opus 4.5. It also achieved 76.8% on SWE-bench Verified, cementing its status as a top-tier coding model, though GPT-5.2 and Opus 4.5 overtake it here at 80 and 80.9, respectively. Moonshot said in a press release sent to reporters that more and more developers have opted to use its offerings, seeing a 170% increase in users between September and November for Kimi K2 and Kimi K2 Thinking, which was released in early November. Agent Swarm and built-in orchestration A key differentiator Moonshot aims to leverage centers on self-directed agents and the Agent Swarm paradigm built into Kimi K2.5. Agentic Swarm has been touted as the next frontier in enterprise AI development and agent-based systems. It has attracted significant attention in the past few months. This capability was highlighted in a new video from VentureBeat collaborator Sam Witteveen, co-founder of Red Dragon AI, a machine learning and training firm in Singapore, on YouTube. For enterprises, this means that if they build agent ecosystems with Kimi K2.5, they can expect to scale more efficiently. But instead of scaling "up" or growing model sizes to create larger agents, it's betting on making more agents that can essentially orchestrate themselves. Kimi K2.5 "creates and coordinates a swarm of specialized agents working in parallel." The company compared it to a beehive where each agent performs a task while contributing to a common goal. The model learns to self-direct up to 100 sub-agents and can execute parallel workflows of up to 1,500 tool calls. "Benchmarks only tell half the story. Moonshot AI believes AGI should ultimately be evaluated by its ability to complete real-world tasks efficiently under real-world time constraints. The real metric they care about is: how much of your day did AI actually give back to you? Running in parallel substantially reduces the time needed for a complex task -- tasks that required days of work now can be accomplished in minutes," the company said. Enterprises considering their orchestration strategies have begun looking at agentic platforms where agents communicate and pass off tasks, rather than following a rigid orchestration framework that dictates when an action is completed. While Kimi K2.5 may offer a compelling option for organizations that want to use this form of orchestration, some may feel more comfortable avoiding agent-based orchestration baked into the model and instead using a different platform to differentiate the model training from the agentic task. This is because enterprises often want more flexibility in which models make up their agents, so they can build an ecosystem of agents that tap LLMs that work best for specific actions. Some agent platforms, such as Salesforce, AWS Bedrock, and IBM, offer separate observability, management, and monitoring tools that help users orchestrate AI agents built with different models and enable them to work together. Multimodal coding and visual debugging Kimi K2.5 also excels in coding and claims to be "the strongest open-source model to date for coding with vision." The model lets users code visual layouts, including user interfaces and interactions. It reasons over images and videos to understand tasks encoded in visual inputs. For example, K2.5 can reconstruct a website's code simply by analyzing a video recording of the site in action, translating visual cues into interactive layouts and animations. "Interfaces, layouts, and interactions that are difficult to describe precisely in language can be communicated through screenshots or screen recordings, which the model can interpret and turn into fully functional websites. This enables a new class of vibe coding experiences," Moonshot said. This capability is integrated into Kimi Code, a new terminal-based tool that works with IDEs like VSCode and Cursor. It supports "autonomous visual debugging," where the model visually inspects its own output -- such as a rendered webpage -- references documentation, and iterates on the code to fix layout shifts or aesthetic errors without human intervention. Unlike other multimodal models that can create and understand images, Kimi K2.5 can build frontend interactions for websites with visuals, not just the code behind them. API pricing Moonshot AI has aggressively priced the K2.5 API to compete with major US labs, offering significant reductions compared to its previous K2 Turbo model. * Input: $0.60 per million tokens (a 47.8% decrease). * Cached Input: $0.10 per million tokens (a 33.3% decrease). * Output: $3.00 per million tokens (a 62.5% decrease). The low cost of cached inputs ($0.10/M tokens) is particularly relevant for the "Agent Swarm" features, which often require maintaining large context windows across multiple sub-agents and extensive tool usage. Modified MIT license While Kimi K2.5 is open-sourced, it is released under a Modified MIT License that includes a specific clause targeting "hyperscale" commercial users. The license grants standard permissions to use, copy, modify, and sell the software. However, it stipulates that if the software or any derivative work is used for a commercial product or service that has more than 100 million monthly active users (MAU) or more than $20 million USD in monthly revenue, the entity must prominently display "Kimi K2.5" on the user interface. This clause ensures that while the model remains free and open for the vast majority of the developer community and startups, major tech giants cannot white-label Moonshot's technology without providing visible attribution. It's not full "open source" but it is better than Meta's similar Llama Licensing terms for its "open source" family of models, which required those companies with 700 million or more monthly users to obtain a special enterprise license from the company. What it means for modern enterprise AI builders For the practitioners defining the modern AI stack -- from LLM decision-makers optimizing deployment cycles to AI orchestration leaders setting up agents and AI-powered automated business processes -- Kimi K2.5 represents a fundamental shift in leverage. By embedding swarm orchestration directly into the model, Moonshot AI effectively hands these resource-constrained builders a synthetic workforce, allowing a single engineer to direct a hundred autonomous sub-agents as easily as a single prompt. This "scale-out" architecture directly addresses data decisionmakers' dilemma of balancing complex pipelines with limited headcount, while the slashed pricing structure transforms high-context data processing from a budget-breaking luxury into a routine commodity. Ultimately, K2.5 suggests a future where the primary constraint on an engineering team is no longer the number of hands on keyboards, but the ability of its leaders to choreograph a swarm.

Moonshot AI releases open-source Kimi K2.5 model with 1T parameters

Moonshot AI releases open-source Kimi K2.5 model with 1T parameters Chinese artificial intelligence developer Moonshot AI today debuted Kimi K2.5, an open-source model that it says can outperform GPT-5.2 across several benchmarks. The launch comes a few days after word emerged that the company is raising capital at a $4.8 billion valuation. Moonshot reportedly closed a separate $500 million round in December. Kimi K2.5 is derived from a large language model called Kimi K2-Base that the company released in early November. One of the latter model's flagship features is that it uses an algorithm called Muon to speed up training. Muon boosts performance by accelerating an LLM's hidden layers, the modules that perform the bulk of the calculations involved in answering prompts. According to Moonshot, its engineers enhanced Kimi K2-Base by training it on 15 trillion tokens' worth of data. The dataset included not only text but also multimodal files. As a result, Kimi K2.5 is better than its predecessor at processing multimodal files such as charts. Moonshot says that the model features a mixture-of-experts architecture with 1 trillion parameters. Those parameters are organized into multiple neural networks that are each optimized for a different set of tasks. When Kimi K2.5 receives a prompt, it doesn't activate all its parameters but only the specific neural network that is best equipped to generate an answer. The result is a significant reduction in hardware usage. The neural networks that make up Kimi K2.5 each include about 32 billion parameters. They're supported by a so-called vision encoder with 400 million parameters. According to Kimi, it's responsible for translating multimodal data uploaded by users into embeddings. Those are abstract mathematical representations that Kimi K2.5's artificial neurons can understand. LLMs use a mechanism called attention to review the data at their disposal and find the details that are most relevant to the task at hand. According to Moonshot, Kimi K2.5 parallelizes the calculations that its attention mechanism uses to identify relevant details. That approach boosts performance because performing calculations side-by-side is faster than completing them one after another. Kimi has a standard mode and a Thinking mode that offers higher output quality. Additionally, a capability called K2.5 Agent Swarm enables the LLM to split complex tasks into simpler sub-steps and assign each sub-step to a separate AI agent. A built-in orchestration engine can create and manage up to 100 agents per prompt. The Kimi K2.5 Agent Swarm serves a similar purpose as the model's parallelized attention mechanism. Agents can perform sub-steps concurrently instead of one after one another to reduce waiting times. Moonshot compared Kimi K2.5 against GPT-5.2, Claude 4.5 Opus and other reasoning models across more than two dozen benchmarks. The company says that its model achieved the highest score on HLE-Full, one of the industry's most difficult LLM evaluations. It comprises 2,500 questions spanning fields such as math and physics. In most of the other benchmarks, Kimi K2.5 came within a few percentage points of the other LLMs' scores. It bested GPT-5.2 on several occasions. Moonshot has made the Kimi K2.5's code available on Hugging Face.

Kimi K2.5 Agent Swarm : Spread Complex Jobs Across 100 Agents, Attack Tasks in Packs

What if your AI could think like a hive mind, tackling complex problems with the precision of 100 synchronized agents? In this guide, Sam Witteveen explains how Kimi K2.5's new Agent Swarm system is reshaping the landscape of artificial intelligence. Unlike traditional models that handle tasks sequentially, Kimi K2.5 uses a network of autonomous sub-agents working in parallel to deliver faster, more accurate results. This innovative approach doesn't just enhance efficiency, it redefines what AI can achieve in fields as diverse as content creation, academic research, and enterprise solutions. Through this breakdown, you'll uncover how Kimi K2.5's modular architecture and Parallel Agent Reinforcement Learning (PAL) enable it to outperform even the most advanced competitors. From its ability to process multimodal inputs like text, images, and videos to its multilingual mastery, Kimi K2.5 is more than just an AI, it's a dynamic ecosystem of specialized capabilities. Whether you're curious about its real-world applications or intrigued by its open source adaptability, this exploration will leave you questioning the limits of what AI can accomplish. Kimi K2.5 is built on a modular architecture that incorporates specialized sub-models, each tailored to specific tasks. These sub-models include: Trained on an extensive dataset of 15 trillion tokens encompassing text, images, and videos, Kimi K2.5 employs reinforcement learning (RL) to adapt to a wide range of challenges. This ensures precise and efficient performance across both technical and creative domains, making it a highly adaptable and reliable AI solution. The Agent Swarm system is the defining feature of Kimi K2.5, allowing up to 100 autonomous sub-agents to collaborate on tasks simultaneously. These sub-agents operate under the coordination of a master orchestrator agent, which assigns tools and manages workflows. By using Parallel Agent Reinforcement Learning (PAL), the system achieves exceptional efficiency in multi-step task execution. This innovative approach allows Kimi K2.5 to outperform competitors such as OpenAI's GPT models and Google's Gemini in both speed and accuracy. The Agent Swarm system is particularly effective in scenarios requiring high levels of parallelization and precision. By distributing tasks among multiple sub-agents, it ensures that even the most complex challenges are addressed efficiently, making it a standout feature in the AI landscape. Master Kimi K2.5 with the help of our in-depth articles and helpful guides. Kimi K2.5 offers a range of advanced features designed to meet the needs of various users, from developers to enterprises. Key capabilities include: These capabilities make Kimi K2.5 a powerful tool for a wide range of applications, from software development to multilingual content creation. Kimi K2.5 is designed to address a broad spectrum of use cases, making it a valuable asset for individuals and organizations across various industries. Its applications include: This versatility ensures that Kimi K2.5 can adapt to the unique requirements of different sectors, from academia to enterprise-level operations. Kimi K2.5 is an open source model, allowing users to download and modify its weights for customized applications. With 1 trillion parameters and 32 billion active parameters at any given time, the model delivers exceptional performance across a wide range of tasks. However, it requires significant GPU resources to operate optimally, reflecting its focus on handling demanding computational workloads. The model's hardware optimization ensures that it can process complex tasks with ease, making it a reliable choice for users with access to high-performance computing resources. This combination of power and flexibility positions Kimi K2.5 as a leading solution in the AI domain. The release of Kimi K2.5 represents a pivotal moment in the evolution of artificial intelligence. Its innovative approach to parallel task execution and advanced verification systems sets a new standard for efficiency and adaptability. By allowing seamless collaboration among autonomous sub-agents, Kimi K2.5 redefines what is possible in AI-driven workflows. As developers and organizations explore its potential, Kimi K2.5 is poised to drive innovation across industries, from software engineering to academic research and beyond. Its open source nature and modular design ensure that it can be tailored to meet the specific needs of users, paving the way for future advancements in AI technology.