Kimi K2.6, Kimi Work agent, Kimi chatbot.
Kimi K2.6 (1T MoE, 32B active, 262K context) posted 80.2% on SWE-bench Verified, 58.6% on SWE-bench Pro (above GPT-5.4's 57.7 and Opus 4.6's 53.4), and a field-leading 54.0 on Humanity's Last Exam with tools, with independent Artificial Analysis serving benchmarks validating real-world deployability. Counterpoint: strength is concentrated in coding/agentic tasks, several headline numbers are vendor-reported, and general multimodal breadth trails Qwen and ERNIE.— Claude Fable 5
Open-sourcing a 1T-parameter frontier model with attention optimizations, a 300-sub-agent Agent Swarm running 4,000 coordinated steps, and the differentiated local-first WebBridge/Kimi Work architecture (data never leaves device) show genuine design originality. Counterpoint: publication footprint and breadth of research areas remain narrower than DeepSeek's or Alibaba's.— Claude Fable 5
Moonshot is expanding from chatbot to platform fast: Kimi WebBridge (May 15) fills the browser-agent gap with local session control supporting Claude Code and Cursor, and Kimi Work (June 8) bundles Agent Swarm, Code, Sheets, Slides, and Deep Research into a productivity suite. Counterpoint: Kimi Work is in limited testing, enterprise-grade tooling and compliance are unproven, and the ecosystem is months old.— Claude Fable 5
ARR topped $200M in April after doubling from $100M in early March, with a ~$2B Meituan-led round at $20B (May 7) and talks for up to $2B more at $30B (June 8), plus a cleared VIE-dismantling path to a Hong Kong IPO. Counterpoint: the implied ~150x price-to-sales multiple is frothy, and as a private company its disclosed financials are limited, warranting some conservatism.— Claude Fable 5
Back-to-back raises are explicitly earmarked for training and inference capacity, and Western providers (CoreWeave #1 on K2.6 price-performance, Cloudflare Workers AI) serve its models at scale. Counterpoint: Moonshot owns no chips, cloud, or disclosed flagship cluster — the weakest infrastructure position in this cohort, renting essentially everything.— Claude Fable 5
WebBridge's local-first design keeping user data on-device is a real privacy-architecture positive, and open weights allow external evaluation. Counterpoint: Moonshot publishes almost no alignment research or deployment-safety documentation, and shipping consumer agents that drive logged-in browser sessions creates substantial unaddressed misuse surface.— Claude Fable 5
Founder Yang Zhilin (Transformer-XL/XLNet co-author) leads one of China's most respected young teams, and the fresh war chest funds hiring against bigger rivals. Counterpoint: the organization is small for the multi-front product expansion it is attempting, and three financings in six months suggest cash intensity that pressures focus.— Claude Fable 5
K2.6 is the #2 most-used LLM on OpenRouter and top-three globally by token usage, with Western channels via CoreWeave and Cloudflare plus Meituan and China Mobile as domestic distribution partners. Counterpoint: cumulative download/install base trails Qwen by an order of magnitude, and the consumer Kimi app's reach is largely China-bound.— Claude Fable 5
Strong open-source mindshare, Meituan/China Mobile strategic channels, and the Kimi consumer usage loop provide emerging flywheels. Counterpoint: open weights mean rivals can host K2.6 without paying Moonshot, and it lacks the proprietary data assets of Alibaba's commerce or Baidu's search.— Claude Fable 5
The strongest momentum in this cohort: ARR doubled in roughly two months, valuation moved $20B to $30B in talks within a month, K2.6 hit #2 on OpenRouter, and WebBridge plus Kimi Work shipped within four weeks of each other. Counterpoint: sustaining this pace requires continuous capital infusions, and the product blitz is ahead of proven enterprise revenue durability.— Claude Fable 5
Tracked for AI Power Rankings scoring. Covers model releases, benchmarks, pricing, funding, partnerships, infrastructure, and policy changes.
Published: 2026-06-08 | Logged: 2026-06-09T07:22Z | Area: Business & Market, Compute & Infra
Kimi developer Moonshot AI is in early talks for a new funding round of more than $1B — as much as $2B — that would value the company at about $30B, its third financing in roughly six months (up from the ~$20B valuation in its early-May raise). The rapid back-to-back rounds reflect the capital intensity of keeping pace in China's intensifying AI race, where rivals are racing to fund compute and model development.
Scoring impact: Positive for Business & Market (50% valuation step-up in ~one month signals strong investor demand) and Compute & Infra (fresh capital to fund training and inference capacity).
Sources:
Published: 2026-06-08 | Logged: 2026-06-11T00:00Z | Area: Product & Platform, Distribution & Reach, Research & Innovation
Moonshot AI has launched Kimi Work, a general-purpose AI agent built on its K2.6 model that runs locally on a user's machine, acting on real files and driving the user's logged-in browser via Kimi WebBridge rather than executing in a cloud sandbox. The integrated workspace bundles Agent Swarm (up to 300 sub-agents), Kimi Code, Sheets, Slides, and Deep Research, and is positioned for knowledge workers handling repetitive multi-tool tasks. The product entered availability this month (initially internal/limited testing).
Scoring impact: Positive for Product & Platform (expands Moonshot from model + chatbot into a full agentic productivity suite). Distribution & Reach lift via a packaged knowledge-worker offering. Research & Innovation credit for the local-first agent architecture leveraging WebBridge.
Sources:
Published: 2026-06-08 | Logged: 2026-06-11T00:00Z | Area: Distribution & Reach, Business & Market, Momentum & Execution
Reporting around Moonshot's new funding talks confirms the K2.6 series now ranks as the second most-used large language model on OpenRouter, and that Moonshot's annual recurring revenue topped $200M in April after roughly doubling from $100M at the start of March. The figures place Moonshot in a four-company front rank of Chinese AI labs (alongside DeepSeek, Zhipu, and MiniMax) collectively seeking valuations exceeding $180B, though Moonshot's ~150x price-to-sales implied multiple is steep.
Scoring impact: Positive for Distribution & Reach (#2 global usage on OpenRouter) and Momentum & Execution (ARR doubling month-over-month). Business & Market context is mixed — strong revenue growth but frothy valuation relative to sales.
Sources:
Published: 2026-05-20 | Logged: 2026-05-21T09:00Z | Area: Business & Market
Moonshot AI has informed shareholders that it intends to dismantle its offshore (VIE) corporate framework, clearing a regulatory path toward an IPO — most likely in Hong Kong under China's new IPO rules. The move follows the company's ~$2B funding round at a $20B valuation led by Meituan's Long-Z, and Kimi's annual recurring revenue topping $200M in April. Restructuring offshore entities is typically a precondition for listing Chinese AI champions on domestic exchanges.
Scoring impact: Positive for Business & Market (clear path to capital markets, signals state-aligned positioning). Mild risk if restructuring complicates foreign investor terms.
Sources:
Published: 2026-05-15 | Logged: 2026-05-16T09:00Z | Area: Product & Platform, Research & Innovation, Safety & Alignment
Moonshot AI released Kimi WebBridge, a Chrome browser extension that lets AI agents simulate human browser operations — logging in, clicking, scrolling, filling forms, scraping content — using the user's existing local session, cookies, and account state. Distinct from the cloud-hosted "computer use" offerings from Anthropic and OpenAI, WebBridge's local-first architecture keeps user data on-device. The plugin advertises support for Kimi Code, Claude Code, and Cursor, targeting use cases from automated information aggregation to quantitative finance backtesting and cross-site price comparison.
Scoring impact: Product & Platform expansion (agent browser-control fills a key gap vs. Anthropic/OpenAI). Research & Innovation lift (local-first architecture is a differentiated design choice). Slight Safety & Alignment positive (on-device data handling reduces some privacy surface).
Sources:
Published: 2026-05-13 | Logged: 2026-05-14T09:00Z | Area: Business & Market, Distribution & Reach
Global Mofy AI Limited announced strategic participation in Moonshot AI's new financing round, citing alignment with Kimi's roadmap in AI-generated content, intelligent production workflows, and digital asset commercialization. The round follows Moonshot's reported ~$2B raise at a $20B valuation in early May and continues to broaden the cap table beyond Chinese mega-cap backers.
Scoring impact: Modest positive for Business & Market (capital, validation) and Distribution & Reach (content/AIGC partnership channel).
Sources:
Area: Model Quality, Compute & Infra, Distribution & Reach
CoreWeave announced that in an independent benchmark by Artificial Analysis covering 11 inference providers running Moonshot's Kimi K2.6, it simultaneously delivered the highest output speed at the most cost-efficient price point. Kimi K2.6 (the 1T-parameter open-source model Moonshot released April 20) is the model being measured, and the result underscores Moonshot's traction as the leading open-source frontier model partner for Western infrastructure providers. The benchmark gives enterprises a credible price-performance reference point for deploying Kimi at scale outside China.
Scoring impact: Positive for Model Quality (Kimi K2.6 validated by independent benchmarks) and Distribution & Reach (CoreWeave as a flagship Western deployment channel). Reinforces open-source moat under Data & Moats.
Sources:
Published: 2026-05-07 | Logged: 2026-05-08T09:00Z | Area: Business & Market, Talent & Org, Distribution & Reach
Moonshot AI, the developer of the Kimi chatbot, is raising approximately $2 billion at a valuation of more than $20 billion, in a round led by Long-Z Investments (Meituan's venture arm) with participation from China Mobile. The company's annual recurring revenue exceeded $200 million in April 2026. Founded by Yang Zhilin in 2023, Moonshot is navigating Beijing's new listing rules for overseas-registered companies as it eyes a public offering. Following its late-April release of the open-source Kimi K2.6 model, Kimi now ranks among the world's top three AI models by token usage on OpenRouter.
Scoring impact: Major positive for Business & Market (capital + valuation step-up cement Moonshot among the elite Chinese AI labs). Strong Distribution & Reach via the Meituan / China Mobile commercial channels. Talent & Org boost from war-chest for hiring. Reinforces standing on global usage charts.
Sources:
Published: 2026-04-20 | Logged: 2026-05-02T09:00Z | Area: Model Quality, Research & Innovation
Moonshot AI released and open-sourced Kimi K2.6 on April 20, a 1T parameter MoE model with 32B active parameters, 262K context window, vision, and agentic capabilities. K2.6 scored 80.2% on SWE-Bench Verified and 58.6% on SWE-Bench Pro (ahead of GPT-5.4 at 57.7, Claude Opus 4.6 at 53.4). On Humanity's Last Exam with tools: 54.0, leading all models. K2.6 can scale its Agent Swarm to 300 sub-agents running 4,000 coordinated steps. The model can execute complex engineering tasks for hours without human intervention. Moonshot is also considering an IPO on the Hong Kong Stock Exchange.
Scoring impact: Major boost to Model Quality (frontier-level SWE-bench scores). Positive for Research & Innovation (open-source 1T parameter model with strong agentic capabilities).
Sources: