MiniMax

Shanghai-based multimodal AI lab spun out of SenseTime — went public on HKSE Jan 2026 as the second Chinese "AI Tiger" to list, with a broad product stack spanning LLMs, video generation, speech, and AI companions.

1. Core Product / Service

MiniMax Group (稀宇科技; SEHK: 100) was founded in December 2021 by former SenseTime computer vision researchers. The company builds a multimodal AI stack:

• MiniMax M-series LLMs — The flagship text model line. MiniMax-M2.7 (released March 2026) is the current top-tier, positioned as a cost-performance play between DeepSeek V4 Flash and V4 Pro. Earlier models: M1 (June 2025), M2.5 / M2.5-Lightning (Feb 2026). The models are colloquially referred to as "Mimo" (迷墨) in Chinese developer circles, particularly Mimo-v2.5 which uses a 60-layer Sliding Window Attention (SWA) architecture optimized for long-text inference [local: 2026-05-31-ai-infrastructure.md].
• Hailuo AI (海螺AI) — Text-to-video generation platform (launched Sep 2024), plus audio functions (Jan 2025). Competes with OpenAI Sora, Kling (Kuaishou), and Jimeng (ByteDance). Hailuo 2.3 is the latest video generation model.
• MiniMax Speech 2.8 — Text-to-speech model supporting 30+ languages (released Apr 2025).
• MiniMax Music 2.5+ — AI music generation.
• MiniMax Agent — Agent framework for autonomous task execution.
• Talkie / Xingye (星野) — AI character companion apps. Talkie (international, launched Jun 2023) hit ~11M MAU in the US by mid-2024, ranking among top-downloaded free entertainment apps [1]. Xingye is the domestic Chinese version.

API access: minimax.io platform, plus third-party distribution via openrouter and DeepInfra.

2. Target Users & Pain Points

• Developers seeking cost-efficient frontier models — M2.7 is priced between DeepSeek Flash and Pro tiers, targeting users who want near-frontier quality at below-frontier prices.
• Content creators / media — Hailuo AI for video generation, Speech for dubbing/localization, Music for production.
• Consumer AI companion users — Talkie/Xingye target the AI girlfriend/boyfriend/character market (similar to character-ai).

Pain points: MiniMax offers the broadest multimodal product stack among Chinese AI labs (text + video + speech + music + companion apps), making it a one-stop-shop vs. the text-centric competition.

3. Competitive Landscape

MiniMax's differentiation: multimodal breadth — it's the only Chinese AI Tiger with competitive products across text, video, speech, music, and consumer companion apps. This breadth makes it structurally different from DeepSeek (text-specialist) and Kimi (agent-specialist).

Pricing positioning (M2.7): targets the gap between DeepSeek V4 Flash (cheapest frontier) and V4 Pro (highest quality), positioning as "best value per quality unit." As of May 2026, DeepSeek and MiniMax/Mimo pricing have essentially converged within the same price band [local: 2026-05-29-summary.md].

4. Unique Observations

• Mimo-v2.5 SWA optimization: The M2.5 model series uses a 60-layer Sliding Window Attention architecture computing only 128-token windows. This makes long-text prefill computation equivalent to traditional 10-layer global GQA, dramatically reducing KV cache memory pressure. Similar SWA techniques are used in Gemma3 and Qwen3, where they auto-activate transparently to users [local: 2026-05-30-summary.md].

• Second Chinese AI lab to go public: MiniMax listed on HKSE on January 9, 2026 — just one day after zhipu (Jan 8, 2026). Together they became the first public-market windows into Chinese AI lab economics. MiniMax's IPO was reported as "shining on debut" [2], but both stocks reflect the tension between Chinese AI policy optimism and US export control headwinds.

• Copyright and distillation controversies: In September 2025, Disney, Universal, and Warner Bros. filed a copyright lawsuit alleging Hailuo AI reproduced copyrighted characters [1]. In February 2026, Anthropic accused MiniMax (alongside two other Chinese AI companies) of using thousands of fraudulent accounts to generate 16M+ interactions with Claude for "distillation" — training their own models on Claude outputs [1]. These legal risks are material for a public company and distinguish MiniMax from labs that haven't faced similar accusations.

• Mimo vs MiniMax naming: The model series is officially named "MiniMax M-series" but is widely referred to as "Mimo" in Chinese developer discourse. Pricing comparisons (e.g., "Mimo vs DeepSeek pricing basically equal") refer to the same MiniMax M2.x model family.

5. Financials / Funding

• Total raised pre-IPO: $600M+ disclosed.
• IPO: Listed on Hong Kong Stock Exchange January 9, 2026, under ticker SEHK: 100 [1][2].
• Employees: 200+ (2025) [1].
• Revenue: not publicly broken out; consumer apps (Talkie) provide in-app purchase revenue; API (minimax.io) provides enterprise revenue.

6. People & Relationships

• Founder / CEO: Yan Junjie (闫俊杰) — ex-SenseTime computer vision researcher.
• Co-founders: Yang Bin (杨斌), Zhou Yucong (周宇聪) — also ex-SenseTime.
• Key investors: Alibaba Group (lead, $600M round), MiHoYo (seed), Hillhouse Investment, HongShan (Sequoia China), IDG Capital, Tencent.
• Listed entity: MiniMax Group Inc., SEHK: 100.
• Competitors: deepseek, kimi, qwen, zhipu.
• Distribution partners: openrouter, DeepInfra.

Sources

[1] Wikipedia, "MiniMax Group," https://en.wikipedia.org/wiki/MiniMax_Group (2026-05-31) [2] South China Morning Post, "Chinese AI start-up MiniMax shines on Hong Kong IPO debut," 2026-01-09, https://www.scmp.com/business/article/3294173/chinese-ai-start-minimax-shines-hong-kong-ipo-debut (2026-05-31) [3] Bloomberg, "Alibaba Backs $2.5 Billion AI Firm in Second Big 2024 Deal," 2024-03-05, https://www.bloomberg.com/news/articles/2024-03-05/alibaba-backs-2-5-billion-ai-firm-in-second-big-2024-deal (2026-05-31)