MiniMax

  Executive Briefing

MiniMax (稀宇科技) is a Shanghai-based frontier AI laboratory and, since January 2026, a publicly listed company (SEHK:0100) that develops multimodal foundation models and consumer AI products. Founded in December 2021 by alumni of SenseTime — China's foremost computer vision company — MiniMax has grown from a stealth start-up into one of the country's five leading "AI Tigers," alongside DeepSeek, Baichuan, Zhipu AI, and Moonshot AI. It is the first Chinese AI start-up to reach a Hong Kong public listing, raising HK$4.82 billion (roughly $618 million USD) in an IPO whose shares more than doubled on their first trading day, implying a market capitalization of approximately $13.7 billion at debut.¹

The lab is led by Yan Junjie (Founder, Chairman, CEO, and CTO), a 36-year-old AI researcher with a PhD from the Chinese Academy of Sciences and more than six years at SenseTime as VP and vice-head of its research institute. Alongside him are co-founders Yun Yeyi (COO), Zhou Yucong (Visual Model Research Leader), and Yang Bin (machine learning research background from Uber ATG), plus Zhao Pengyu, who was appointed Executive Director and LLM Research Leader in June 2025. The team's founding thesis was that consumer AI products and frontier foundation models should be built simultaneously — that products would generate the diverse training data required for genuine general intelligence — a philosophy Yan rooted in his conviction, formed after reading the GPT-3 paper in 2020, that multimodal general intelligence was achievable through engineering and scale.

MiniMax is best known globally for three things: Talkie, an international AI companion app that reached 29 million monthly active users by the end of 2024 and ranked fourth among the most-downloaded AI apps in the United States in the first half of that year; Hailuo AI, a text-to-video and multimodal creation platform that directly competed with OpenAI Sora and accumulated more than 600 million cumulative video generations; and the MiniMax M-series open-weight frontier models, culminating in MiniMax-M3 (June 2026), which claimed 59.0% on SWE-Bench Pro with a 1-million-token native context window. The company's stated mission is building general-purpose AI for the public, with an explicit long-term target of achieving AGI, and it has staked out an unusually global revenue posture — more than 70% of its 2025 revenue came from outside China, with Singapore and the United States as its top markets.²

  At a Glance

  Origins & Founding

MiniMax was founded in December 2021 in Shanghai by four colleagues who had spent years at SenseTime, China's leading AI and computer vision company. Yan Junjie, who became CEO and CTO, had spent more than six years there as VP and vice-head of the research institute, overseeing deep-learning technologies and distributed computing systems. His interest in language models crystallized in 2019 when OpenAI's bots defeated elite human teams in Dota 2, and he became convinced after reading the GPT-3 paper in 2020 that multimodal general intelligence was achievable by applying engineering rigor and scaling laws — not mysticism. He later articulated this as a founding principle: "AI is not mysticism. It is an engineering problem that can be decomposed using first principles."⁵

The co-founders brought complementary depth. Yun Yeyi (COO, age 31 at founding) had led fundraising, strategy, and innovative business development at SenseTime from 2017 to 2022 before taking a bachelor's degree covering electrical engineering, economics, and mathematics from Johns Hopkins University. Zhou Yucong led algorithms R&D at SenseTime and subsequently worked in algorithm development at Huawei Technology before co-founding MiniMax. Yang Bin came from a different trajectory: he had co-founded the research center for Uber Advanced Technologies Group (Uber ATG, later sold to Aurora Innovation) and subsequently worked for autonomous-driving startup Waabi, bringing machine-learning systems expertise rooted at the University of Toronto.⁶

The founding thesis was deliberately unconventional for 2021: rather than building a foundation model lab first and productizing later, MiniMax committed from day one to developing both foundational models and consumer-facing applications in parallel. Yan's argument was that consumer products would generate the variety and scale of training data required for genuine general intelligence, and that real-world deployment feedback would accelerate research faster than academic benchmarks alone. The company secured early angel funding of approximately $30 million from Yunqi Capital, IDG Capital, Hillhouse Capital, and a notable strategic backer: MiHoYo, the Shanghai studio behind Genshin Impact, which recognized gaming as an AI-native domain and provided early credibility at a reported post-money valuation of approximately $200 million.⁷

  History & Timeline

    2021–2022: Stealth founding and the Glow experiment

In its first year, MiniMax operated largely out of public view, building its foundational model stack. In October 2022 it launched Glow, an AI companion chatbot app in China, which reached approximately five million users within four months — a striking early signal of consumer demand for AI characters. The rapid growth masked a content moderation problem: a product manager later disclosed that roughly 80% of Glow's users were generating borderline or explicit content with AI characters, a disclosure made public when Glow was removed from Chinese app stores in March 2023 without official explanation from the company.⁸

    2023: Expansion, Series A, and international launch

MiniMax used the period after Glow's removal to broaden its approach. In June 2023 it launched Talkie on Google Play and Apple's App Store — an international AI character chat application styled after the Glow concept but targeting a global audience, including avatars of public figures such as Taylor Swift and Elon Musk. That same month a Series A financing of more than $250 million, led by a Tencent-backed entity, closed at an undisclosed valuation.⁹ In September 2023 MiniMax launched Xingye (Xing Ye) in China as the domestic successor to Glow, establishing a dual-track consumer strategy: Talkie for international markets, Xingye for China.

    2024: Hailuo AI, Series B, and multimodal expansion

March 2024 saw two simultaneous moves: the launch of Hailuo AI, a multimodal creation platform, and the close of a $600 million Series B led by Alibaba Group at a reported valuation of $2.5 billion, with participation from Hillhouse, HongShan, and IDG Capital.¹⁰ Talkie continued to scale, reaching approximately 11 million monthly active users by July 2024 and 29 million by December 2024, placing it fourth among the most-downloaded AI apps in the United States in the first half of the year.² In September 2024, MiniMax released the Video-01 text-to-video model and Music-01 music generation model under the Hailuo AI platform, entering direct competition with OpenAI Sora and Runway.

    2025: Open-weight models, IPO, and geopolitical friction

January 2025 marked MiniMax's arrival as a recognized frontier open-weight lab: it released and open-sourced the MiniMax-01 series — MiniMax-Text-01 and MiniMax-VL-01 — a 456-billion-parameter model family with a 4-million-token context window, accompanied by a technical paper on arXiv.¹¹ A confidential IPO filing was submitted to the Hong Kong Stock Exchange in July 2025, targeting a valuation above $4 billion. That same month, NVIDIA CEO Jensen Huang held a reported two-hour private meeting with Yan Junjie in Beijing and publicly praised MiniMax as "world-class." A Series B extension of $300 million at a $4 billion valuation, led by Shanghai STVC Group (state-owned capital), followed.⁴

The year also brought escalating geopolitical and legal friction. MiniMax was added to the U.S. Department of Commerce Entity List in 2025, requiring export licenses for specified technologies amid allegations of ties to China's military-industrial complex. In September 2025, Disney, Universal, and Warner Bros. Discovery filed a reported $75 million copyright infringement lawsuit alleging that Hailuo AI had generated videos and images of copyrighted characters without authorization.¹² June 2025 saw a major product release event — MiniMaxWeek — at which the company launched MiniMax M1, Hailuo 02 (1080p video generation), MiniMax Agent, Speech-02, and Hailuo Video Agent.¹³

    2026: IPO, self-improving models, and continued frontier releases

On 9 January 2026, MiniMax Group Inc. listed on the Hong Kong Stock Exchange (ticker: SEHK:0100), raising HK$4.82 billion (approximately $618–619 million USD). Shares were priced at HK$165 and closed their first trading day at HK$345, a gain of 109%, implying a market capitalization of roughly $13.7 billion at debut and making Yan Junjie a billionaire at age 36.¹ Cornerstone investors in the IPO included Alibaba, the Abu Dhabi Investment Authority (ADIA), Boyu Capital, and Mirae Asset Securities; CICC and UBS served as joint sponsors.

In February 2026, Anthropic publicly accused MiniMax — alongside two other unnamed Chinese firms — of operating fraudulent accounts that generated more than 16 million interactions with Claude AI to harvest model outputs for training, a form of model distillation the company characterized as theft.¹⁴ MiniMax had not publicly responded to these allegations as of the latest available sources. The same month saw the release of MiniMax-M2.5 and M2.5-Lightning, which reportedly exceeded 3.07 trillion token calls in their first week and became the first Chinese model on OpenRouter to surpass 50 billion tokens in daily consumption. MiniMax-M2.7 followed on 18 March 2026 as the first model to autonomously run more than 100 reinforcement-learning training iterations, analyzing failure trajectories and modifying its own scaffold code.¹⁵ In April 2026, a partnership with Volcano Engine (ByteDance's enterprise software unit) was announced to co-build an AI software platform. The MiniMax-M3 open-weight frontier model arrived on 1 June 2026, reporting 59.0% on SWE-Bench Pro with a 1-million-token context window and native multimodal image and video understanding.¹⁶

  Mission, Philosophy & Research Agenda

MiniMax's stated mission is to build general-purpose AI for the public, with an explicit long-term target of achieving AGI. CEO Yan Junjie frames this not as a moonshot but as a tractable engineering program: "AI is not mysticism. It is an engineering problem that can be decomposed using first principles."⁵ He has explicitly rejected the "Chinese OpenAI" label, positioning MiniMax as a global LLM provider whose primary constituency is individual consumers — two-thirds of its 2025 revenue came from individual users — rather than a state-adjacent enterprise AI player.²

Three research priorities define the lab's distinctive approach. First, long-context efficiency: MiniMax pioneered the Lightning Attention mechanism, a linear-complexity attention architecture that achieves near-linear computational scaling with sequence length, enabling the 4-million-token context of MiniMax-01 and the 1-million-token native context of M1 and M3 without proportional compute cost. This work is closely tied to the constraint of GPU export restrictions: building efficient architectures is partly a strategic response to the hardware disadvantage created by U.S. trade controls. Second, full-modality integration from day one: unlike labs that began with text and added modalities later, MiniMax targeted text, vision, video, speech, and music simultaneously from its 2021 founding, believing that multimodal diversity of training signals would accelerate progress toward general intelligence. Third, self-improving agentic models: M2.7's autonomous RL training loops — in which the model analyzed its own failure trajectories and modified its scaffold code across more than 100 iterations — represent MiniMax's most publicly visible bet on models that improve themselves, a research direction the lab intends to extend through the M-series roadmap.

Yan acknowledges a capability gap versus leading U.S. labs but argues that constraint itself drives innovation: "The best models in the United States are stronger than the best models in China. That's a fact. But the difference is not as large as people think."⁵

  Research & Publications

MiniMax has published foundational research on long-context architectures and reasoning models, with most of its key results available as open arXiv preprints and open-weight model releases on GitHub.

• MiniMax-01: Scaling Foundation Models with Lightning Attention (arXiv:2501.08313, January 14, 2025)¹¹ — introduced the Lightning Attention mechanism combining linear and softmax attention in a 7:1 ratio across 80 transformer layers; 456B total / 45.9B activated parameters; 4M-token context window; matched GPT-4o and Claude 3.5 Sonnet performance on standard benchmarks while providing a 20–32× longer context window. The paper listed 90 authors.
• MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention (arXiv:2506.13585, June 16, 2025)¹³ — described the first open-weight large-scale hybrid-attention reasoning model; native 1M-token context (claimed to be 8× that of DeepSeek R1); MoE combined with Lightning Attention; reportedly reduced inference compute by 75% versus comparable models.
• Lightning Attention / Hybrid Attention kernel research — additional arXiv publications on GPU kernel implementation (arXiv:2510.21956) and distilling softmax-attention models into hybrid attention via KL-guided layer selection (arXiv:2512.20569).
• MiniMax-M2.7 self-evolution methodology (March 2026)¹⁵ — the first documented production model to autonomously run more than 100 reinforcement-learning training iterations without human intervention, analyzing failure trajectories and modifying its own training scaffold.
• MiniMax-M3 MSA architecture (June 2026)¹⁶ — introduced Modified Sparse Attention achieving 20× lower per-token compute at 1M context, 9× faster prefill, and 15.6× faster decoding versus the prior generation; 59.0% SWE-Bench Pro (company-reported, not independently verified as of this writing).

Open-weight model weights, code, and associated documentation are available at github.com/MiniMax-AI, covering MiniMax-01, MiniMax-M1, and subsequent M-series releases.

  Models & Products

MiniMax ships through a dual surface: developer and enterprise APIs and a set of consumer apps that together reach hundreds of millions of cumulative users across more than 200 countries. Its model generations have progressed from early proprietary ABAB models through the landmark MiniMax-01 series and into the rapidly iterated M-series.

Foundation models:

• MiniMax-Text-01 — 456B parameter (45.9B activated) LLM with Lightning Attention and MoE architecture; 4M-token context; open-sourced January 2025.
• MiniMax-VL-01 — Companion multimodal vision-language model to Text-01; open-sourced January 2025.
• MiniMax M1 — Open-weight hybrid-attention reasoning model; 1M-token native context; released June 2025; described in arXiv:2506.13585.
• MiniMax-M2.5 / M2.5-Lightning — Released February 2026; exceeded 3.07 trillion token calls in its first week (reported).
• MiniMax-M2.7 — Released March 18, 2026; self-improving agent model; first to run 100+ autonomous RL training iterations; 56.22% SWE-Pro (reported).
• MiniMax-M3 — Released June 1, 2026; open-weight frontier; 1M-token native context; 59.0% SWE-Bench Pro (company-reported, unverified independently); native multimodal image and video understanding.

Media and speech models:

• Hailuo AI / Video-01 — Text-to-video generation platform launched March 2024; Video-01 model released September 2024; Hailuo 02 (June 2025) reaches 1080p at up to 10 seconds; 600M+ cumulative video generations.
• Music-01 — AI music generation model; released September 2024.
• Speech-02 — Text-to-speech model supporting 30+ languages; released April 2025; ranked first on TTS benchmarks (company-reported).

Consumer applications:

• Talkie — International AI character and companion chat app (iOS and Android); launched June 2023; 29M monthly active users globally by December 2024; ranked fourth most-downloaded AI app in the U.S. in H1 2024.
• Xingye (Xing Ye) — Domestic Chinese AI companion community app; launched September 2023; successor to the original Glow app.
• MiniMax Agent — General-purpose AI assistant for document analysis, presentation creation, and agentic workflows; launched June 2025.
• Hailuo Video Agent — Video creation tool with templates for content creators; launched June 2025.

➡ See individual model cards for specifications, pricing, and benchmark results.

  People

MiniMax's leadership is tight-knit by design, reflecting the founding cohort's shared background and the company's small headcount (415 employees as of 2025). Yan Junjie (age 36) holds the combined Chairman, CEO, and CTO roles — an unusual configuration signaling that research leadership is not separated from commercial execution. He holds a PhD in AI from the Chinese Academy of Sciences (2015) and completed postdoctoral work at Tsinghua University; his academic output spans more than 200 publications with over 30,000 citations and includes the Wu Wenjun AI Award (2019) and the Guangdong Technology Invention Award (2020).⁶

Yun Yeyi (age 31) serves as COO and Executive Director; she holds a BA covering electrical engineering, economics, and mathematics from Johns Hopkins University and spent five years in fundraising and strategy roles at SenseTime before co-founding MiniMax. Zhou Yucong (age 32) is Executive Director and Visual Model Research Leader, with a BS in Mathematics and MS in Computer Science from Beihang University, and prior experience in algorithm development at SenseTime and Huawei. Yang Bin, the fourth co-founder, brings machine learning research depth from his tenure co-founding Uber ATG's research center (subsequently sold to Aurora Innovation) and work at autonomous-driving startup Waabi; his current operational title within MiniMax is not clearly documented in public filings.¹⁷ Zhao Pengyu (age 29) was appointed Executive Director and LLM Research Leader in June 2025; he holds a BS and MS in Computer Science from Peking University and worked in recommendation algorithm research at Beijing Hulu Technology from 2020 to 2023.

The lab's alumni network is still nascent given its youth, but the founders' own SenseTime diaspora origin is the more significant genealogical fact: MiniMax is the most prominent AI laboratory to emerge from SenseTime's research alumni.

  Funding, Ownership & Business

MiniMax raised approximately $1.15 billion in pre-IPO financing across four rounds before its Hong Kong listing, assembling a shareholder base that spans strategic Chinese technology companies, global institutional investors, and Chinese state capital. The financing history is as follows (all figures reported):

MiniMax's business model operates across two segments: AI-native consumer products (Talkie, Hailuo AI, MiniMax Agent — subscriptions and advertising, representing 67% of 2025 revenue at approximately $53.1 million) and Open Platform / enterprise API (the remaining 33%, approximately $26 million, from 214,000 enterprise customers and developers).³ Full-year 2025 revenue reached $79.0 million, up 158.9% year over year, though losses remain substantial: the company reported a net deficit estimated at approximately $512 million for the first nine months of 2025 alone, reflecting a research and infrastructure spend that dwarfs current revenue by roughly a 9:1 ratio.³ More than 70% of revenue originates from outside China, positioning MiniMax as the most internationally-oriented of China's AI Tigers. The company's reported market capitalization as of June 11, 2026 was approximately HK$293–309 billion (roughly $37–40 billion USD), a figure that is fast-moving and should be treated as a reported estimate.⁴

  Partnerships & Ecosystem

MiniMax's investor roster doubles as its strategic partnership network, with Alibaba and Tencent holding stakes that imply potential cloud and distribution relationships. MiHoYo — the studio behind Genshin Impact — was an early strategic backer whose involvement signaled AI's relevance to gaming and provided credibility in China's technology ecosystem.⁷ Alibaba Group led the $600 million Series B and served as a cornerstone investor in the IPO, with an implied cloud and compute relationship, though no formal cloud exclusivity has been publicly disclosed.¹⁰ Tencent led or co-led the Series A, extending an ongoing strategic investor relationship.

On the state side, Shanghai STVC Group, a Shanghai government-linked investment vehicle, led the July 2025 Series B extension, signaling alignment with Shanghai's ambitions to be a center of China's AI industry. The Abu Dhabi Investment Authority (ADIA) served as a cornerstone IPO investor, a notable signal of Middle Eastern sovereign-wealth interest in Chinese frontier AI.

In April 2026, MiniMax announced a strategic partnership with Volcano Engine (ByteDance's enterprise software unit) to co-build a software platform intended to enhance AI capabilities across enterprise workflows.⁴ While NVIDIA and CEO Jensen Huang are not formal investment partners, Huang's reported two-hour private meeting with Yan Junjie in Beijing in July 2025 — followed by a public endorsement of MiniMax as "world-class" — underscored NVIDIA's strategic interest in China's frontier AI ecosystem, even as U.S. export restrictions constrain the GPU supply available to MiniMax.⁵

  Compute & Infrastructure

Compute access is the central infrastructure constraint for MiniMax, shaped in large part by the U.S. export restrictions on advanced NVIDIA GPUs that limit the hardware available to Chinese AI labs. MiniMax's technical response has been architectural: the Lightning Attention mechanism is explicitly designed to achieve near-linear computational complexity for long-context processing, enabling competitive context windows at lower hardware cost than pure softmax-attention systems. The M3 architecture's Modified Sparse Attention (MSA) extends this, claiming 20× lower per-token compute at 1M context compared to the prior generation.¹⁶

The lab has not publicly disclosed its data-center footprint, training cluster specifications, or cloud provider relationships in detail. Its IPO prospectus stated an intention to use proceeds for foundation-model research over the following five years, which would include infrastructure investment. Compute constraints contributed to MiniMax's emphasis on inference efficiency — the M1 paper reported a 75% reduction in inference compute versus comparable models — as a competitive differentiator within the Chinese AI ecosystem.¹³

  Notable Events & Controversies

MiniMax has navigated a series of significant events since its founding that illuminate both its capabilities and the risks of operating at the intersection of consumer AI, geopolitics, and frontier model development.

• Glow content moderation incident (March 2023) — MiniMax's first consumer app, Glow, was removed from Chinese app stores without official explanation. A product manager later disclosed that approximately 80% of Glow's users were generating borderline or explicit adult content with AI characters, raising questions about the company's moderation infrastructure that it has not publicly addressed in detail.⁸

• U.S. Entity List addition (2025) — MiniMax was added to the U.S. Department of Commerce Entity List during 2025 amid concerns about alleged ties to China's military-industrial complex. The specific month of addition and the precise allegations have not been confirmed across sources. The listing requires U.S. companies to obtain export licenses for specified technologies sold or transferred to MiniMax, constraining its access to advanced AI hardware and potentially U.S.-headquartered cloud and software services.⁹

• Disney / Universal / WBD copyright lawsuit (September 2025) — The Walt Disney Company, Universal, and Warner Bros. Discovery filed a reported $75 million copyright infringement lawsuit in a U.S. court against MiniMax, alleging that the Hailuo AI video generation platform had produced videos and images of copyrighted characters without authorization. The suit is ongoing and has not been adjudicated as of this writing.¹²

• Anthropic distillation accusation (February 2026) — Anthropic publicly accused MiniMax, alongside two other Chinese firms it did not name, of operating fraudulent accounts that generated more than 16 million interactions with the Claude AI to harvest model outputs for training — a form of model distillation that Anthropic characterized as model theft. MiniMax had not publicly responded to these allegations as of available sources; the accusations have not been independently verified or adjudicated.¹⁴

• Hong Kong IPO debut (January 9, 2026) — MiniMax's shares surged 109% on their first trading day (HK$165 offer price to HK$345 close), making it the best-performing major Chinese tech IPO in years and marking Yan Junjie as a billionaire at 36. The debut established MiniMax as the first Chinese AI start-up to achieve a Hong Kong public listing, preceding Zhipu AI's own listing by a short interval.¹

  Competitive Position

Within China, MiniMax sits among the five "AI Tigers" alongside DeepSeek, Baichuan, Zhipu AI, and Moonshot AI (Kimi). Its differentiation within this cohort is meaningful on several axes: it is the only member with large-scale international consumer app distribution (Talkie at 29M MAU globally); it is the most internationally revenue-oriented (70%+ of 2025 revenue from outside China); and it was the first to list publicly on a major exchange. The M-series model cadence — from MiniMax-01 in January 2025 through M3 in June 2026 — has established a pace of open-weight releases that keeps MiniMax visible on global leaderboards.

Globally, Hailuo AI competes directly with OpenAI Sora, Runway, and Kuaishou's Kling in text-to-video generation; Talkie competes with Character.AI in AI companion applications; and the M-series open-weight models sit on the same competitive leaderboards as DeepSeek R1, GPT-4o, and Claude 3.5 Sonnet. MiniMax's M3 claimed to surpass GPT-5.5 and Gemini 3.1 Pro on SWE-Bench Pro at 59.0%, though this figure is from company-released materials and has not been independently verified as of this writing.¹⁶

MiniMax's competitive liabilities are its heavy R&D burn rate (~9:1 ratio of spending to revenue), its geopolitical exposure (U.S. Entity List status, GPU access constraints), ongoing copyright litigation, and the structural disadvantage of competing with better-capitalized rivals including Alibaba Qwen, ByteDance, and Baidu domestically, and OpenAI, Anthropic, and Google DeepMind internationally.

  Outlook & Roadmap

MiniMax has stated that IPO proceeds will be directed toward foundation-model research over the next five years, with an R&D focus that spans model capability, inference efficiency, and consumer product development.¹ The strategic roadmap announced through mid-2026 centers on five areas: continuing the M-series model cadence with increasingly capable open-weight reasoning and agentic models; expanding Hailuo AI's video generation capabilities (Hailuo 02 already reaches 1080p at up to 10 seconds); deepening the Talkie and Xingye consumer moats globally; growing the MiniMax Agent platform for enterprise agentic workflows; and scaling the developer and API ecosystem, which had reached 214,000 enterprise customers and developers by the time of its IPO filing.

The lab's three core technical bets — self-improving models (M2.7's autonomous RL methodology), long-context efficiency (M3's 1M-token architecture), and full multimodal integration — are expected to converge in future M-series releases. Strategic risks include the ongoing U.S. Entity List status and GPU export restrictions (which constrain hardware access and may affect cloud and software relationships), the unresolved Disney / Universal / Warner Bros. Discovery copyright lawsuit, the significant gap between R&D expenditure and revenue, and intensifying competition both domestically and internationally. The reported market capitalization of roughly $37–40 billion as of June 2026 implies substantial growth expectations that are not yet reflected in current revenues.⁴

---

  References

1. MiniMax Hong Kong IPO — Nikkei Asia
2. MiniMax full-year 2025 financial results — PR Newswire
3. MiniMax 2025 financials — PR Newswire
4. MiniMax company overview — Wikipedia
5. Meet MiniMax — Asia Tech Lens
6. MiniMax management — investor relations
7. MiniMax deep dive — Turing Post
8. MiniMax / Glow content issues — Rest of World
9. MiniMax — Wikipedia
10. MiniMax Series B — Wikipedia
11. MiniMax-01 paper — arXiv:2501.08313
12. MiniMax-M1 paper — arXiv:2506.13585
13. Disney / WBD lawsuit — Wikipedia
14. Anthropic distillation accusation — Wikipedia
15. MiniMax-M2.7 announcement
16. MiniMax-M3 / June 2026 releases — Wikipedia
17. Yang Bin role uncertainty — multiple sources reviewed; current title not confirmed in public filings

  References

1. MiniMax's January 9, 2026 Hong Kong IPO — raised HK$4.82B (~$618-619M USD); offer price HK$165; day-one close HK$345 (+109%); implied market cap ~$13.7B — Nikkei Asia. ↩ ↩² ↩³ ↩⁴

2. MiniMax global revenue split and Talkie MAU — MiniMax 2025 full-year results, PR Newswire. ↩ ↩² ↩³

3. Full-year 2025 revenue ($79.0M, +158.9% YoY) and segment breakdown — PR Newswire; net loss figure for first 9 months of 2025 (~$512M) is a reported estimate from IPO prospectus disclosures and may not reflect full-year results. ↩ ↩² ↩³

4. Market cap (~$37–40B as of June 11, 2026 at ~HK$435/share), ByteDance / Volcano Engine partnership, and Series B extension — figures are fast-moving and reported; Wikipedia: MiniMax. ↩ ↩² ↩³ ↩⁴ ↩⁵

5. Yan Junjie's philosophy and Jensen Huang meeting — Asia Tech Lens profile. ↩ ↩² ↩³ ↩⁴

6. Founding team biographies — MiniMax investor relations; Turing Post. ↩ ↩²

7. MiHoYo early investment and angel round — Turing Post; valuation ~$200M post-money is reported. ↩ ↩²

8. Glow removal and content moderation context — Rest of World; MiniMax provided no official explanation. ↩ ↩²

9. Series A ($250M+, Tencent-backed lead) and Entity List — Wikipedia: MiniMax; Series A lead investor identity and exact amount have slight variation across sources. ↩ ↩²

10. Series B ($600M, Alibaba lead, $2.5B valuation) — Wikipedia: MiniMax; figures reported. ↩ ↩²

11. MiniMax-01 technical paper — arXiv:2501.08313; MiniMax blog. ↩ ↩²

12. Disney / Universal / Warner Bros. Discovery copyright lawsuit (Sept 2025, reported $75M) — Wikipedia: MiniMax; unresolved as of this writing. ↩ ↩²

13. MiniMax-M1 technical paper and MiniMaxWeek announcements — arXiv:2506.13585. ↩ ↩² ↩³

14. Anthropic's February 2026 accusation of 16M+ Claude interactions for model distillation — Wikipedia: MiniMax; unverified and unadjudicated; MiniMax has not publicly responded. ↩ ↩²

15. MiniMax-M2.7 self-evolution announcement — MiniMax blog. ↩ ↩²

16. MiniMax-M3 benchmark claims (59.0% SWE-Bench Pro) are from company-released materials and have not been independently verified as of this writing — Wikipedia: MiniMax. ↩ ↩² ↩³ ↩⁴

17. Yang Bin's current operational title is not clearly documented in MiniMax's public IPO filings or investor relations materials; his co-founder status is confirmed across sources. ↩