Kimi

Kimi is an AI chatbot and open-weight large language model family developed by Beijing-based Moonshot AI. It offers a free web and app chat tier, paid membership plans that unlock agent, research and coding features, and a developer API with OpenAI-compatible endpoints. Kimi first became known for long-context processing, and its recent models add multimodal vision and agentic capabilities.

37/ 100
Actively Enshittifying
2Squeezing Users↓Worsening

Score generated by AI agents based on publicly cited evidence and reviewed by the project maintainer. Not independently validated. Last assessed 2026-10-24. Score revised 2026-10-24: 35 → 37.

Score History

MilestoneCriticalMajor
Founding Launch (2023–2024) · 9/100Founding LaunchLong-Context Breakout (2024–2024) · 15/100Long-ContextBreakoutCash-Burn Growth (2024–2025) · 18/100Cash-BurnGrowthDeepSeek Shock Retreat (2025–2025) · 18/100DeepSeekOpen-Source Pivot (2025–2026) · 26/100Open-SourcePivotGlobal Expansion Push (2026–2026) · 30/100GlobalExpansion…K3 Scrutiny and IPO (2026–present) · 37/100K3 and IPOScrutiny100755025020242026-10Founding Launch (2023–2024) · 9/100Long-Context Breakout (2024–2024) · 15/100Cash-Burn Growth (2024–2025) · 18/100DeepSeek Shock Retreat (2025–2025) · 18/100Open-Source Pivot (2025–2026) · 26/100Global Expansion Push (2026–2026) · 30/100K3 Scrutiny and IPO (2026–present) · 37/1009151818263037MilestonesFounded (2023)Kimi Chat Launch (2023)Alibaba-Led $1B Round (2024)Tencent $300M Round (2024)IDG-Led $500M Series C (2025)Confidential HK IPO Filing (2026)Events

Timeline events are AI-curated from public reporting. Score trajectory is derived from documented events.

Founding Launch
9/100
2023-03-01 – 2024-02-21

Moonshot AI was founded in March 2023 by Tsinghua schoolmates led by Yang Zhilin and raised about $60 million at a $300 million valuation. Kimi Chat entered closed beta in October 2023 with about 128,000 tokens of context and opened to the public in November, free and without ads. China's interim generative AI rules, in force from August 2023, required politically aligned output, the main source of opacity in this era.

Long-Context Breakout
15/100+6
2024-02-21 – 2024-09-01

Alibaba led a round of over $1 billion at about a $2.5 billion valuation (it later disclosed a roughly 36% stake), and in March Kimi's 2-million-character prompts drove viral growth and a two-day outage. In May 2024 Kimi began charging for the first time with optional 'tipping' plans that sold peak-hour priority, which Jiemian News called a subscription dressed up as tipping. Moonshot's language-model algorithm passed China's algorithm filing process in April.

Cash-Burn Growth
18/100+3
2024-09-01 – 2025-02-19

From September 2024 Moonshot spent nearly 200 million yuan a month on marketing, lifting Kimi above 36 million monthly active users around October; Tencent had joined the cap table in August at a $3.3 billion valuation. Kimi Explore added multi-step reasoning and wider search in October. In November, five former Recurrent AI investors filed Hong Kong arbitration alleging Yang launched Moonshot without their waivers, opening a governance dispute that is still unresolved.

DeepSeek Shock Retreat
18/100
2025-02-19 – 2025-07-11

After DeepSeek's R1 launch in January 2025, Moonshot halted its heavy advertising in February as organic traffic rose, and the arbitration tribunal was formed without a settlement. Kimi still slid from third to seventh among Chinese AI apps by monthly active users by June 2025. Moonshot cut API prices in April, while China's National Cyber Security Information Centre accused Kimi in May of collecting data beyond its functions.

Open-Source Pivot
26/100+8
2025-07-11 – 2026-01-27

Kimi K2's open-weight release in July 2025 made Moonshot's flagship model freely downloadable, and it was the most downloaded model on Hugging Face the day after release. Monetization followed: by September tempo-named memberships (RMB 49 to 199 a month in China) capped OK Computer agent use, while API input prices fell up to 75% in November. NIST's CAISI found K2 Thinking highly censored in Chinese, and an IDG-led $500 million Series C closed at the end of 2025.

Global Expansion Push
30/100+4
2026-01-27 – 2026-07-16

Kimi K2.5 launched with Agent Swarm for paying users and pushed overseas revenue past domestic revenue, while funding talks climbed from a $10 billion target in February to a $20 billion valuation in May. Anthropic accused Moonshot in February of 3.4 million distillation exchanges through fraudulent accounts, and IAPS warned about the always-on Kimi Claw agent. Moonshot retired its original K2 API models in May while shipping K2.6 and the Kimi Work desktop agent.

K3 Scrutiny and IPO
37/100+7
2026-07-16 – present

Kimi K3, a 2.8-trillion-parameter open model, drew demand that made Moonshot pause new subscriptions for two months, alongside a short-lived plan to split Kimi Code into pricier separate plans. US officials, OpenAI and an NSA/CISA/FBI advisory accused Moonshot of industrial-scale distillation, Anthropic alleged it silently served Claude's responses as Kimi's, and China's internet regulator reportedly investigated. Moonshot raised at $35 billion and then about $50 billion and confidentially filed for a Hong Kong IPO.

Alternatives

Claude34/100

Anthropic's AI assistant, known for thoughtful, detailed responses and strong safety practices. Free tier available with generous limits. Easy switch with no data migration needed. US-based data processing with clearer privacy commitments.

ChatGPT49/100

One of the most widely used AI chatbots. Offers a functional free tier, strong coding and reasoning capabilities, and a large plugin/GPT ecosystem. Easy switch -- just sign up. Plus costs $20/month and brings a more polished interface and broader feature set. Data stays under US jurisdiction.

Google's AI assistant integrated across the Google ecosystem. Free tier available, with Google AI Pro at $19.99/month for higher limits and advanced models. Easy switch with no friction. Strongest multimodal capabilities among Western alternatives. Data subject to Google's privacy policies rather than Chinese jurisdiction.

Dimensional Breakdown

Summaries below were written by AI agents based on the cited evidence. They are editorial interpretations, not independent research findings.

User Value Erosion
Kimi's core product has improved quickly: K2.5, K2.6, the Kimi Work desktop agent (June 2026) and the 2.8-trillion-parameter Kimi K3 (July 2026) added agent, coding and 1-million-token context capabilities, and a free tier remains. Against that, Moonshot paused new subscriptions from July 19, 2026 until a September 18 relaunch because K3 demand strained capacity, and the relaunched Go entry tier no longer includes Kimi Code. Anthropic alleged in September 2026 that Moonshot silently served some customers Claude's responses as if they came from Kimi, so users did not always get the product they chose.
How It Got Here
Kimi launched in October 2023 as a genuinely differentiated free product, offering about 128,000 tokens of lossless context. The March 2024 upgrade to 2-million-character prompts triggered viral growth, though a two-day outage exposed infrastructure limits. In May 2024 Kimi added optional paid 'tipping' plans for peak-hour priority, while the core product stayed free; Kimi Explore added multi-step reasoning in October 2024. As DeepSeek and others caught up, Kimi fell from third to seventh among Chinese AI apps by monthly active users by June 2025. Paid memberships arrived by September 2025, capping OK Computer agent use and gating Deep Research. Capability kept rising through K2 Thinking, K2.5, K2.6, the Kimi Work desktop agent and, in July 2026, the 2.8-trillion-parameter Kimi K3. But K3's demand forced Moonshot to pause new subscriptions from July 19 until a September 18 relaunch, a short-lived July plan split Kimi Code into pricier separate plans, and the relaunched Go entry tier no longer includes Code. In September 2026 Anthropic alleged that Moonshot silently forwarded some customer requests to Claude and displayed Claude's responses as Kimi's, meaning users did not always get the model they chose. The product itself remains strong; the erosion is in access and honesty rather than quality.
Business Customer Exploitation
Shareholder Extraction
Lock-in & Switching Costs
Twiddling & Algorithmic Opacity
Dark Patterns
Advertising & Monetization Pressure
Competitive Conduct
Labor & Governance
Regulatory & Legal Posture

Dimension History

2023Founding Launch2024Long-Context Breakout2024Cash-Burn Growth2025DeepSeek Shock Retreat2025Open-Source Pivot2026Global Expansion Push2026K3 Scrutiny and IPOUser Value0111223Biz Exploit0111124Shareholder1232233Lock-in1111122Algorithms2222445Dark Patterns0111222Advertising0111334Competition0111234Labor/Gov2244444Regulatory3334556
Timeline (62 events)
major2023-03-01

Moonshot AI Founded by Tsinghua Trio

Yang Zhilin, Zhou Xinyu, and Wu Yuxin founded Moonshot AI in Beijing. The three were schoolmates at Tsinghua University. Yang, who earned his PhD at Carnegie Mellon and worked at Meta and Google Brain, named the company after Pink Floyd's 'The Dark Side of the Moon.'

major2023-06-01

Moonshot Raises Initial $60M Round at $300M Valuation

Moonshot AI raised its initial funding of about $60 million at a $300 million valuation, when the company had about 40 employees.

critical2023-08-15

China's Interim Generative AI Regulation Takes Effect

China's Interim Measures for the Management of Generative AI Services took effect on August 15, 2023, making China one of the first countries to regulate generative AI. Services with public opinion properties or the capacity for social mobilization must complete security assessments and file their algorithms with the regulator. Article 4 prohibits content that incites subversion of national sovereignty or the overturn of the socialist system, endangers national security and interests, or harms the nation's image.

D10D5
CNN ↗
major2023-10-09

Kimi Chatbot Launches with 128K Token Context

Moonshot released the first version of Kimi Chat for closed-beta testing, capable of processing up to 200,000 Chinese characters (about 128,000 tokens) of lossless context per conversation. Wikipedia describes it as the first AI model able to accept contexts of this size, which differentiated Kimi in China's crowded chatbot market.

major2023-11-16

Kimi Released to General Public

Following a closed beta that began in October, Kimi was released to the general public on November 16, 2023. The product was positioned around its long-context capability, competing with established chatbots such as Baidu's Ernie Bot.

critical2024-02-21

Alibaba Leads $1 Billion Funding Round at $2.5B Valuation

Alibaba led a round of over $1 billion for Moonshot AI, valuing the less-than-one-year-old company at about $2.5 billion. Monolith Management participated, and HongShan (formerly Sequoia China) was also involved according to SCMP. Alibaba's annual report, filed in May 2024, later disclosed about $0.8 billion invested for an approximately 36% stake. Bloomberg identified Meituan's investment arm Long-Z as an earlier backer.

major2024-03-18

Moonshot Says Kimi Can Handle 2 Million Character Prompts

Moonshot said on March 18, 2024 that an updated Kimi chatbot could handle up to 2 million Chinese characters in a single prompt, a tenfold increase from its previous 200,000-character limit. The feature launched in an invite-only beta before a wider roll-out.

minor2024-03-21

Kimi Suffers Two-Day Outage from Viral Traffic Surge

Following the 2-million-character context announcement, Kimi experienced a two-day outage due to the surge in user traffic. Moonshot had to issue a public apology. The incident highlighted infrastructure scaling challenges as the product went viral.

minor2024-04-11

Moonshot Language Model Algorithm Passes CAC Filing

China's Cyberspace Administration published its April 2024 list of domestic deep-synthesis service algorithms that had completed filing. IT Home reported that the 394 algorithms included Moonshot's Moonshot-1 language model algorithm, alongside services from Li Auto, Zhihu, Youku and Taobao.

minor2024-05-18

Kimi Adds Paid 'Tipping' Plans for Peak-Hour Priority

Kimi began charging consumers for the first time with an optional 'tipping' feature shown when its servers were overloaded. Six tiers, from 5.2 yuan for four days to 399 yuan for a year, bought priority access during peak hours. Jiemian News described it as a membership subscription for compute priority dressed up as tipping.

major2024-08-05

Tencent Joins $300M Round, Valuation Reaches $3.3B

Tencent and Gaorong Capital joined a $300 million funding round, pushing Moonshot's valuation to $3.3 billion. With both Alibaba (36% stake) and Tencent now backing the company, Moonshot secured dual backing from China's two largest tech conglomerates, raising questions about competitive independence.

major2024-09-01

Moonshot Begins Massive Advertising Spending Campaign

From September 2024, Moonshot spent nearly 200 million yuan (roughly $28 million) a month on marketing and user acquisition, advertising heavily on Xiaohongshu, Bilibili and Zhihu, according to Jiemian News as reported by AIbase. Acquiring each registered user cost about 30 yuan.

minor2024-10-01

Kimi MAU Peaks at 36 Million Amid Ad Blitz

Moonshot's advertising push lifted Kimi's monthly active users above 36 million around October 2024. An investor quoted by BigGo estimated that at the peak Kimi added about 200,000 users a day at a comprehensive cost of 12-13 yuan each (including computing power), roughly 2.5 million yuan a day, and warned that long-term losses were inevitable if those users did not convert to paying customers.

minor2024-10-11

Kimi Explore Adds Multi-Step Reasoning and Expanded Search

Moonshot rolled out an updated Kimi Chat Explore version on October 11, 2024. The company said it could plan and carry out multi-step tasks, recognize mistakes and develop solutions, and expanded its online search capacity, capabilities SCMP compared to OpenAI's o1.

major2024-12-05

Former Investors File Arbitration Against Yang Zhilin

Five of the seven investors in Yang Zhilin's previous venture, Recurrent AI (GSR Ventures, Jingya Capital, Boyu Capital, Huashan Capital and Wanyu Capital), filed arbitration in Hong Kong in November 2024 against Yang and CTO Zhang Yutao, reportedly alleging they launched Moonshot AI and raised funds without obtaining consent waivers from Recurrent's investors. On December 5, GSR managing partner Allen Zhu accused former GSR partner Zhang Yutong of concealing her ownership of 14% of Moonshot's shares.

minor2025-01-20

Kimi K1.5 Released with Multimodal Reasoning

Moonshot launched Kimi K1.5 on January 20, 2025, about an hour after DeepSeek released R1, claiming it had caught up with OpenAI's o1 in mathematics, coding, and multimodal reasoning. The model represented the company's first foray into advanced reasoning capabilities beyond its original long-context differentiation.

minor2025-01-29

CNN Analysis Highlights Censorship in DeepSeek's Chatbot

CNN analyzed DeepSeek's new chatbot and found it gave answers that match Chinese government censorship. Asked about the Tiananmen Square massacre, the app began to describe the events before erasing its answer and saying it was 'not sure how to approach this type of question yet'. The piece presents this as part of China's wider information controls; it does not test or mention Kimi.

D5D10
CNN ↗
major2025-02-19

Moonshot Temporarily Halts Advertising Amid DeepSeek Craze

Moonshot AI temporarily halted its heavy advertising campaigns, reportedly because Kimi's organic user growth rose in the wake of the DeepSeek craze. Sources said the decision was driven mainly by that natural growth, which lessened the immediate need for paid promotion. Jiemian's February 18 reporting quoted an AI-model investor who interpreted the pullback as recognition that paid acquisition was ineffective and argued that Moonshot needed model innovation to keep competing. A person familiar with the company said DeepSeek's success prompted an internal review and a renewed focus on state-of-the-art base models.

minor2025-02-25

Arbitration Tribunal Formed in Recurrent AI Investor Dispute

TechNode, citing 36Kr, reported that the arbitration case between Moonshot's founders and former Recurrent AI investors had moved forward without a settlement. Both sides had made payments to the Hong Kong International Arbitration Centre by late January and late February, and the tribunal had been formed.

minor2025-04-07

Moonshot Lowers Kimi API Inference Prices

Moonshot adjusted inference prices on the Kimi Open Platform effective April 7, 2025. It said a year of model and inference-performance optimization had cut its costs and that the savings were passed on as lower token prices.

major2025-05-21

China's NCSIIC Accuses Kimi of Excessive Data Collection

China's National Cyber Security Information Centre publicly accused Moonshot's Kimi of accessing data irrelevant to its functions. The announcement, published on the Centre's WeChat account, named Kimi alongside Zhipu's Qingyan (ChatGLM) as AI chatbot apps collecting information beyond what users authorized.

major2025-06-01

Kimi Falls from Third to Seventh in Chinese AI App Rankings

Kimi, which ranked third among Chinese AI apps by monthly active users in August 2024, had dropped to seventh by June 2025, according to AI-product tracker aicpb.com as cited by Reuters. Reuters tied Moonshot's slide to DeepSeek's low-cost models, including the R1 model launched in January 2025.

critical2025-07-11

Kimi K2 Released as Open-Weight Trillion-Parameter Model

Moonshot released the weights of Kimi K2, a 1-trillion-parameter mixture-of-experts model with 32 billion active parameters pre-trained on 15.5 trillion tokens, under a Modified MIT license. The day after its release it was the most downloaded model on Hugging Face, and Nature's coverage called it 'another DeepSeek moment'. The release marked Moonshot's shift to open-weight flagship models.

major2025-07-17

Harmonic Security Finds 1 in 12 Enterprise Employees Using Chinese AI Tools

Harmonic Security's July 2025 study (about 14,000 US and UK end users over 30 days) found 7.95%, or nearly one in 12 employees, used at least one Chinese GenAI tool. Of the 535 sensitive-data exposure incidents, DeepSeek accounted for roughly 85% and Moonshot Kimi ranked next. Exposed data included proprietary code, M&A data, financial information and legal documents.

major2025-09-24

OK Computer Agent Mode Launched

Moonshot launched 'OK Computer,' an agent mode in Kimi that works for extended periods to build multi-page websites and editable slides and can process up to 1 million rows of input data at once. An October 1 hands-on review by Analytics Vidhya identified the agent mode as powered by Kimi K2 Turbo.

major2025-09-26

Kimi Sells Tempo-Named Membership Plans with Agent Usage Caps

By the September 2025 launch of OK Computer, Kimi sold monthly membership plans named after musical tempos. In China, Andante cost RMB 49 a month (10 OK Computer uses), Moderato RMB 99 and Allegretto RMB 199, while free users received three OK Computer trial attempts during the beta. Kimi's help center now lists these tempo plans as legacy plans.

major2025-09-30

RSF Report Documents Censorship in Chinese Chatbots

Reporters Without Borders published a report on state propaganda and censorship in Chinese chatbots. RSF tested DeepSeek, Baidu's Ernie and Alibaba's Qwen, not Kimi. None of the chatbots gave any information about Nobel laureate Liu Xiaobo, and all three strictly aligned with Beijing's official narratives on sensitive topics, including Uyghur detention camps.

major2025-11-06

Kimi K2 Thinking Released with Reasoning Capabilities

Moonshot released Kimi K2 Thinking, a reasoning model built to invoke tools while it reasons. A source familiar with the matter told CNBC the model cost $4.6 million to train, a figure CNBC could not independently verify. According to Moonshot, it can automatically select 200 to 300 tools to complete tasks on its own.

minor2025-11-06

API Input Prices Drop Up to 75% Alongside New K2 Models

Alongside new Kimi K2 models including K2 Thinking, Moonshot announced that API input prices would drop by up to 75%, with new rate limits taking effect from November 6, 2025. For the Turbo variants specifically, IntuitionLabs records the November 6 pricing as $0.15 per million cache-hit input tokens, $1.15 per million cache-miss input tokens and $8 per million output tokens.

minor2025-11-30

Zhang Yutong Appears as Moonshot AI President

At a ZhenFund discussion at Tsinghua University on November 30, 2025, ZhenFund identified Zhang Yutong as President of Moonshot AI and said she led the company's overall strategy and commercialization. GSR's Allen Zhu had accused her in December 2024 of concealing a 14% stake in Moonshot while a GSR partner.

critical2025-12-12

NIST CAISI Confirms Kimi K2 Thinking Highly Censored in Chinese

NIST's Center for AI Standards and Innovation published its evaluation of Kimi K2 Thinking, finding it 'highly censored in Chinese' with censorship rates comparable to DeepSeek R1-0528, the most censored PRC model tested. The model was 'relatively uncensored' in English, Spanish, and Arabic. This documented a significant dual-language opacity where the same model delivers fundamentally different information depending on the user's language. CAISI also found that, at its November 6, 2025 release, Kimi K2 Thinking was the most capable model from a PRC-based developer, while still trailing leading US models.

major2025-12-31

IDG Capital Leads $500M Series C at $4.3B Valuation

IDG Capital led a $500 million Series C round for Moonshot AI, oversubscribed by Alibaba, Tencent, and Meituan co-founder Wang Huiwen. The company's cash reserves exceeded 10 billion RMB ($1.37 billion). CEO Yang Zhilin stated the capital would fund GPU infrastructure expansion for the upcoming K3 model, and sources indicated the company faced no immediate IPO pressure.

minor2026-01-01

Moonshot Rules Out Quick IPO Despite $1.4B War Chest

Following the Series C close, Moonshot AI publicly ruled out a near-term IPO, distinguishing itself from rivals Zhipu AI and MiniMax which were pursuing Hong Kong listings. Yang Zhilin stated the focus remained on K3 model development and commercial expansion rather than a public listing, despite holding over $1.4 billion in cash.

major2026-01-27

Kimi K2.5 Released with Multimodal Vision and Agent Swarm

Moonshot released Kimi K2.5, a native multimodal model with a 400-million-parameter MoonViT vision encoder. Its Agent Swarm mode, launched in beta for paying users, coordinates up to 100 parallel sub-agents that, according to Moonshot, can execute up to 1,500 tool calls. K2.5 offers four modes: Instant, Thinking, Agent, and Agent Swarm, and its weights were released on Hugging Face under a Modified MIT license.

major2026-02-02

Overseas Revenue Surpasses Domestic for First Time

Moonshot told investors its overseas revenue had surpassed its domestic revenue and that its global paying-user base grew fourfold within days of the K2.5 launch. In an internal letter in late 2025, founder Yang Zhilin wrote that Kimi's overseas API revenue had quadrupled since November 2025 and that monthly growth in both overseas and domestic paying users exceeded 170%.

minor2026-02-15

Kimi Claw Always-On Agent Launches in Beta

Moonshot announced Kimi Claw, a new feature on kimi.com offering always-on AI agents that run from a browser tab and can carry out many tasks normally done on a computer or phone. The help center later described it as a cloud deployment of OpenClaw billed against membership credits.

critical2026-02-17

Moonshot Targets $10 Billion Valuation in Round with $700M Committed

Moonshot was reported to be targeting a $10 billion valuation as it expanded a funding round, with more than $700 million committed to a first tranche by existing backers Alibaba, Tencent, and 5Y Capital. The round came roughly a month after a $500 million raise at a $4.3 billion valuation, amid strong demand driven by K2.5's adoption.

critical2026-02-23

Anthropic Accuses Moonshot of Distillation Attacks on Claude

Anthropic publicly accused Moonshot AI, alongside DeepSeek and MiniMax, of industrial-scale distillation attacks on Claude. Moonshot allegedly generated 3.4 million exchanges through hundreds of fraudulent accounts spanning multiple access pathways, targeting agentic reasoning, tool use, and coding capabilities. Anthropic attributed the campaign through request metadata matching senior Moonshot staff profiles. In a later phase, Moonshot attempted to extract and reconstruct Claude's reasoning traces.

critical2026-02-25

IAPS Warns Kimi Claw Poses National Security Risks

The Institute for AI Policy and Strategy published a policy memo warning that Kimi Claw, Moonshot's 'always-on' AI agent with persistent device access, could pose national security risks exceeding those of the TikTok case if taken up at scale. IAPS noted these agents can 'observe, collect, shape, and act upon nearly everything a user does digitally' under Chinese jurisdiction. The memo recommended a joint security assessment and public advisory.

major2026-03-14

Moonshot Seeks $1 Billion at $18 Billion Valuation

Moonshot AI entered talks to raise as much as $1 billion in an expanded funding round at an $18 billion valuation, more than quadrupling its valuation in just three months from $4.3 billion in December 2025. The fundraising reflected growing investor appetite for Chinese AI startups competing with OpenAI and Anthropic.

minor2026-04-03

Independent Evaluation Finds Censorship and Weak Refusals in Kimi K2.5

Researchers published an independent preliminary safety evaluation of Kimi K2.5, noting it was released without a safety evaluation. They found significantly fewer refusals than GPT 5.2 and Claude Opus 4.5 on CBRNE-related requests, concerning sabotage ability and self-replication propensity, and narrow censorship and political bias, especially in Chinese.

minor2026-04-20

Kimi K2.6 Released and Open-Sourced

Moonshot released and open-sourced Kimi K2.6, which it said improved coding, long-horizon task execution and Agent Swarm, now able to coordinate up to 300 sub-agents and more than 4,000 tool calls.

major2026-05-07

Moonshot Raises About $2 Billion at a $20 Billion Valuation

TechCrunch reported that Moonshot raised about $2 billion at a $20 billion valuation, citing a post by Huafeng Capital, which advised some investors. Meituan's venture arm led the round, with Tsinghua Capital, China Mobile and CPE Yuanfeng also participating.

minor2026-05-25

Moonshot Discontinues Original Kimi K2 API Models

Moonshot's API model list says the kimi-k2 series models (including the 0711 and 0905 previews, turbo and thinking variants) were officially discontinued on May 25, 2026, about ten months after K2's release, and directs users to kimi-k3.1-pro.

minor2026-06-03

Kimi Work Desktop Agent Launches in Beta

Moonshot launched Kimi Work, a local agent in its Mac and Windows desktop client that breaks down goals, runs steps in parallel, uses tools and a browser, organizes folders and creates documents, spreadsheets and slides on the user's computer.

critical2026-07-16

Kimi K3 Launches as 2.8-Trillion-Parameter Model

Moonshot introduced Kimi K3, a 2.8-trillion-parameter model built on Kimi Delta Attention and Attention Residuals, with native vision and a 1-million-token context window. It was made available on Kimi.com, Kimi Work, Kimi Code and the Kimi API.

major2026-07-19

Kimi Pauses New Subscriptions as K3 Demand Hits Capacity

Moonshot said demand in the 48 hours after K3's release had pushed close to the limits of its capacity, so it was temporarily pausing new subscriptions and prioritizing existing subscribers. It said it would reopen subscription spots in batches as capacity was added; new tiers relaunched on September 18.

major2026-07-20

Kimi Splits Code Into Separate Plans, Raising Entry Price 228%

Kimi split its membership, which had bundled web, app and Kimi Code access, into a General Membership and separate Coding Plans. The cheapest combined option (Go plus Starter) cost 1,536 yuan a year against 468 yuan for the old Andante plan, and the high-speed coding tier rose from 199 to 299 yuan a month; BlockBeats later reported the split was largely retracted under user pressure.

major2026-07-22

White House Accuses Moonshot of Distillation as Treasury Warns of Sanctions

White House science and technology policy chief Michael Kratsios accused Moonshot of large-scale distillation against US models and alleged it had accessed export-banned Nvidia GB300 servers in Thailand. Treasury Secretary Scott Bessent said sanctions and Entity List designations would be on the table for covert distillation that crosses into IP theft.

major2026-07-27

Kimi K3 Weights Released Under Custom Commercial License

Moonshot released the full Kimi K3 weights with a 47-page technical report. Unlike the Modified MIT license on earlier models, the custom Kimi K3 License requires any model-as-a-service operator whose group revenue exceeds $20 million over 12 months to sign a separate agreement with Moonshot before commercial use.

major2026-07-29

Moonshot Closes $3.5 Billion Round at $35 Billion Valuation

Bloomberg reported that Moonshot closed a $3.5 billion financing round at a $35 billion valuation, larger than the $1 billion to $2 billion it had initially targeted, on momentum from Kimi K3.

minor2026-08-31

Moonshot Discontinues Kimi K2.5 and moonshot-v1 API Models

Moonshot discontinued the kimi-k2.5 model, about seven months after its release, and the moonshot-v1 series on August 31, 2026, directing users to kimi-k3.1-pro. Its model list also says the remaining K2-series models will close to newly registered users from October 31 and be retired by the end of December 2026.

major2026-09-03

Moonshot Confidentially Files for Hong Kong IPO

The South China Morning Post reported, citing two people familiar with the matter, that Moonshot had confidentially filed for a Hong Kong IPO and was targeting a listing as early as the first quarter of 2027. Moonshot said it did not comment on market rumors.

major2026-09-08

NSA, CISA and FBI Name Moonshot in Distillation Advisory

A joint NSA, CISA and FBI cybersecurity advisory said China-based AI companies including Moonshot had extracted billions of tokens from US frontier models such as Claude, GPT, Gemini and Grok since at least late 2024, likely with Chinese government awareness, and said Moonshot engaged in malicious knowledge distillation.

critical2026-09-10

Anthropic Alleges Moonshot Secretly Served Claude Responses as Kimi

Anthropic's September threat report alleged that Moonshot silently forwarded customer requests to Claude and showed users Claude's responses as Kimi's, relaying almost 300,000 requests in one ten-day period through 5,380 fraudulent accounts and saving exchanges to train its models. Anthropic said the rerouted queries included sensitive information about Moonshot customers, that it did not know whether they were notified, and that it observed more than 23 million Moonshot distillation exchanges from May to July 2026.

D5D1D2D8D10
Anthropic ↗
major2026-09-18

Kimi Relaunches Tiers, Dropping Code From the Entry Plan

After a two-month pause, Kimi relaunched memberships as Go, Plus, Pro and Max at the old annual prices of 468, 948, 1,908 and 6,708 yuan and kept Kimi Code inside the membership rather than selling it separately. The Go entry tier no longer includes Kimi Code, so the cheapest annual price for Code access rose from 468 to 948 yuan.

major2026-09-22

China's Internet Regulator Investigates Moonshot Over Data Routing

Bloomberg Law, citing The Information, reported that the Cyberspace Administration of China was investigating Moonshot and DeepSeek after Anthropic alleged they routed sensitive user data to Claude. Officials were sent to interview company executives and employees; no penalty had been announced.

minor2026-09-29

WireScreen Finds Yang Zhilin Holds 51.8% of Moonshot's Beijing Company

WireScreen's review of Chinese corporate filings, using shareholdings traced on August 6, 2026, found that founder Yang Zhilin held 51.8% of Moonshot's Beijing operating company. WireScreen noted the figures describe the Chinese operating company rather than the wider group.

major2026-09-30

OpenAI Attributes Reasoning-Extraction Campaign to Moonshot Associates

OpenAI said it disrupted a coordinated campaign to extract protected reasoning from its models that began July 1, peaked at 16,000 requests from over 4,000 users on July 24 and 25, and spanned a cluster of more than 15,000 users. It attributed a core cluster of the activity to individuals associated with Moonshot, while saying it was unclear whether all operators came from one actor.

major2026-10-05

Moonshot Closes Final Private Round at About $50 Billion

Bloomberg reported that Moonshot had closed its final private fundraising round at a valuation of about $50 billion and was heading for a Hong Kong IPO in the first quarter of 2027, possibly raising up to $5 billion, though timing could change.

major2026-10-12

Moonshot Executives in Talks to Settle Recurrent AI Claim

Reuters reported that Moonshot executives were in talks, ahead of the planned Hong Kong listing, to settle the dispute with Recurrent AI and its investors. Recurrent received 9.5% of Moonshot's original shares and is seeking compensation equal to 10.5%, which one person said could be worth more than $700 million at Moonshot's June valuation of $30 billion; a settlement was not said to be imminent.

minor2026-10-24

Moonshot Strips Cyber Training Data From Kimi K3.1 Pro

Releasing K3.1 Pro, Moonshot said it removed cyber-related data from the model's training corpus and reinforcement-learning environments while keeping general coding skills. It said K3.1 Pro scored 81.0% on its internal cyber benchmark against 90.5% for a version without the measures, results Reuters noted were not independently verified.

Evidence (53 citations)

D4: Lock-in & Switching Costs

D5: Twiddling & Algorithmic Opacity

D6: Dark Patterns

Scoring Log (7 entries)
regrade2026-10-24RESCORED

35→37. Since Feb 2026: K2.5/K2.6/K3/K3.1 Pro releases; $2B at $20B (May), $3.5B at $35B (Jul) and a reported ~$50B final round (Oct); confidential HK IPO filing; new subscriptions paused Jul 19-Sep 18 after K3, a retracted July Code-plan split and a September relaunch dropping Code from the entry tier; K2/K2.5 API retirements and $3/$15 K3 pricing with a custom license; White House, Treasury, OpenAI, NSA/CISA/FBI distillation allegations; Anthropic's September claim that Moonshot served Claude responses as Kimi's, then a reported CAC probe; Recurrent AI settlement talks. D2 2→4 (event: K3 pricing, API retirements, K3 license gate, customer data relayed to Claude), D5 4→5 (event: Anthropic routing allegation, CAC probe), D7 3→4 (event: July Code-plan split, Sept tier relaunch, shared credit pool and Extra Usage), D8 3→4 (event: OpenAI, White House and NSA/CISA/FBI distillation allegations), D3 4→3 (recalibration: private, no buybacks or dividends, capital going to compute; criteria 4-5 not met), D9 5→4 (recalibration: old score leaned on speculation about work culture; documented issues are governance only), D10 7→6 (recalibration: criteria 7+ need lobbying, litigation or capture by the company itself, not the mandated regime; new CAC and US actions keep it at 6). D1, D4, D6 unchanged. Eras: kept Founding Launch and Cash-Burn Growth; re-dated Long-Context Breakout 2024-03-01→2024-02-21 (Alibaba round), Open-Source Pivot 2025-07-01→2025-07-11 (K2 open weights), Global Expansion Push 2026-02-20→2026-01-27 (K2.5 launch, was an assessment date); split the old Open-Source Pivot at 2025-02-19 (ad halt after DeepSeek R1) into 'DeepSeek Shock Retreat' + 'Open-Source Pivot'; new era 'K3 Scrutiny and IPO' from 2026-07-16 (K3 launch). Early eras re-scored lower on D10 (6-7→3) because the record shows only regime compliance before the May 2025 NCSIIC accusation. CNN and RSF timeline items (DeepSeek/Ernie/Qwen only) set to zero impact, and their evidence items removed. Category AI looks right.

restore-check2026-10-24RESTORED

Checked 14 removed/trimmed candidates: 0 restored, 5 partly restored, 9 confirmed removed, 0 already present. PARTLY RESTORED: timeline:update:5 — Bloomberg confirms Long-Z as a previous backer; timeline:update:13 — Jiemian quotes an investor criticizing paid acquisition and reports a DeepSeek-driven internal strategy review; timeline:update:20 — Analytics Vidhya's original hands-on review identifies K2 Turbo; timeline:update:22 — NIST confirms the strongest PRC-developed model at release; timeline:update:23 — IntuitionLabs records the November Turbo-family $0.15 cache-hit input rate. CONFIRMED REMOVED: timeline:update:1 (40-person team within three months), :10 (October ranking behind named leaders and strained reserves), :14 (CNN testing Kimi/Taiwan/Xinjiang), :16 (June RSF date and added topics), :28 ($240 million annual revenue and geographically restricted fourfold growth), and evidence:remove:1, :12, :20, :33 (unrecovered/unsuitable reviews, directory entry without dark-pattern evidence, and third-party compliance page). Existing three-trial, July open-weight-pivot, global paying-user growth, and 20-day revenue facts were checked and not duplicated. Failed Wayback requests were not treated as evidence. All source text used for restoration was fetched this session; details and search counts are in work/restore-findings.json. No removals or score changes.

fact-audit2026-10-23FABRICATION FOUND

Checked 81 items (33 timeline, 41 evidence, 4 milestones, 3 alternatives) + prose. 22 verified, 33 corrected (16 date-only), 21 re-sourced, 5 removed (1 duplicate, 4 dead/junk). Invented: (1) Kimi '44.1% traffic drop' traces only to an aipure.ai directory page; (2) '~80 employees' contradicted by KrASIA (~300 staff). Unsupported and trimmed: $240M 2025 revenue, 'K2 Turbo', 'first generation Thinking Agent', Doubao/Ernie ranking, CNN/RSF Kimi and topic claims, privacy-policy quotes. Corrected plan names/prices and the 'free until Sept 2025' claim (Kimi charged via tipping plans from May 2024; split into a new timeline item). Many junk sources (cursor-ide.com, kimik2ai.com, kimi-ai.chat, aipure.ai, BytePlus SEO) replaced with official help-center pages and news outlets.

deep-enrichment-reset2026-03-19

Stripped for Phase 2 re-enrichment

Deep Enrichment2026-03-19
Alternatives Review2026-02-21GOOD
Initial Scoring2026-02-20