MiniMax Group

MiniMax Group Inc. is a Shanghai-based AI lab, incorporated in the Cayman Islands, that builds foundation models and a run of consumer apps on top of them: the M-series large language models, priced to undercut Western frontier labs on agentic coding and reasoning; Hailuo for video and image generation; and Talkie, a companion-chat app TEXXR’s corpus first flagged in December 2024 as one of the most-downloaded AI apps in the US, going head-to-head with Character.AI out of a Shanghai lab most American users had never heard of.

MiniMax listed on the Hong Kong Stock Exchange on January 9, 2026, ticker 0100.HK, one trading day after fellow Chinese AI lab Zhipu. The two were the first of China’s foundation-model “AI tigers” — the DeepSeek-era cohort chasing OpenAI and Anthropic on capability — to reach a public listing. Both American labs were still private as of this writing: OpenAI and Anthropic each filed confidentially for their own IPOs in June 2026, five months after MiniMax and Zhipu had already listed. MiniMax priced at HK$165 and closed its first session at HK$345, up 109%. A December 2025 filing had targeted a raise near $538.5M against a $4B-plus valuation; priced at the top of an already-raised range, the final deal brought in roughly $619M.

Two structural facts matter more than the debut pop. First, weighted voting rights: founder Yan Junjie holds about 25% of the equity but close to 72% of the vote, through Class B shares worth ten votes each; a second insider adds a further slice, putting combined insider control near 79% of votes on ordinary resolutions. Second, float: only about 18 million shares — roughly 5% of the company — were tradable in the first six months, and the retail tranche was oversubscribed more than a thousand times over. Neither is unusual alone for a Hong Kong tech listing; together, on a stock this thin, they explain the next section. A standard US brokerage account does not reach 0100.HK, either: it trades in Hong Kong dollars on HKEX hours, through a broker with HK market access, under HKEX and Hong Kong Companies Ordinance disclosure rules rather than SEC ones — and there is no US ADR.

TEXXR’s earliest mention of MiniMax runs to May 3, 2024: a Financial Times survey grouping it with Zhipu AI, Moonshot AI, and 01.ai as four Chinese generative-AI startups, together valued between $1.2B and $2.5B, trying to match their American counterparts. Coverage stayed thin and financial-heavy through 2024 — an Alibaba-led funding round, a Wall Street Journal piece on Chinese labs writing more efficient code because they’d been cut off from the top tier of AI chips, a Talkie profile — then built as MiniMax filed confidentially for a Hong Kong listing in July 2025, targeting a valuation above $4B.

The financials disclosed along the way: 2024 revenue of $30.5M, then $79M in 2025, up 159% and ahead of the $71.4M analysts had modeled — against a net loss that widened from $465.2M in 2024 to $1.87B in 2025, the year of the IPO. By May 2026 MiniMax said annualized recurring revenue had reached $300M and filed to prepare a second listing, on Shanghai’s STAR Market — an A+H structure several of the tigers, MiniMax and Zhipu both, are now chasing to tap mainland capital alongside Hong Kong’s.

The stock’s own path is the sharper story. After the 109% debut pop, MiniMax kept climbing on a float thin enough to amplify any demand at all: shares reached roughly HK$840 by late May, and peaked near HK$1,330 in March — better than seven times the offer price — before a hard reversal. On July 9, 2026, a lock-up expiry released 153 million shares, about 49% of the company, into the market; MiniMax fell 18% that day and another 12% the next, part of a slide that had by then erased more than 80% of the March peak, though the stock was still trading well above its HK$165 offer price. MiniMax answered the same week with a roughly $2B financing — new shares plus convertible bonds — and a memo from Yan Junjie promising to forgo his own salary until the company reaches AGI: a confidence signal timed to the lock-up shock, not a change to the loss line above it.

A separate dispute runs alongside the stock chart. In February 2026, Anthropic accused MiniMax, DeepSeek, and Moonshot of violating Claude’s terms of service — running Claude more than 16 million times and using the outputs to train their own models via distillation, echoing an earlier complaint OpenAI made to US lawmakers about DeepSeek. A year before that, Disney, Universal, and Warner Bros. Discovery jointly sued MiniMax, alleging its Hailuo video tool was trained on their copyrighted characters — the kind of claim a foreign listing doesn’t make disappear, since MiniMax holds no meaningful US assets to attach a judgment to. Both disputes sit inside the same constraint: Chinese labs can’t buy Nvidia’s top-end training silicon, and TEXXR’s corpus has tracked that ceiling pushing the whole cohort toward cheaper, more efficient models and outside-the-model shortcuts since at least August 2024.

TEXXR’s knowledge graph carries 59 edges on MiniMax across 26 articles running from that May 2024 mention to July 2026, skewed financial (56% of edges) and launch (29%), with competitor, controversy, legal, acquisition, and regulatory edges splitting a thin remainder. That is a record that treats MiniMax mostly as a capital event and a model-release cadence, not as a fight — the fights, when they show up, show up as a handful of sharp-angle edges rather than a sustained thread.

Coverage volume has a shape worth reading against the stock’s own chart. TEXXR logged one or two MiniMax articles a quarter through 2024 and most of 2025, seven in the final quarter of 2025 as the IPO filing landed, and a peak of 12 in the first quarter of 2026 — the listing quarter. It slowed to nine in the second quarter and stands at four so far in the third, a deceleration TEXXR’s momentum reading scores at -0.56 velocity and -0.31 acceleration. The coverage record cooled at almost exactly the point the stock did, off the same March-to-July peak-to-trough.

One pairing in the graph is worth reading directly. The only edge connecting MiniMax to DeepSeek at all is the February 2026 Anthropic complaint, which names both as accused of the same distillation conduct — despite running in the same tiger cohort under the same compute constraints, the two have no direct competitor edge in TEXXR’s graph. The sharpest competitor framing TEXXR’s graph carries pairs MiniMax with Zhipu, not DeepSeek: a June 2026 JPMorgan note explicitly picked “Zhipu as a winner against rival MiniMax,” and the market agreed for months after — Zhipu’s shares closed up 33% the day the note landed against a comparatively muted 7% for MiniMax, and by June Zhipu traded at a valuation an order of magnitude past its listing-day twin.

Order flow is signed before it becomes revenue — the earliest of the three views. From the capital-flow ledger.

Signals in the coverage record worth tracking, not trade ideas:

• The Shanghai STAR Market filing. MiniMax and Zhipu are both pursuing an A+H structure. Whether the CSRC counseling filing converts to an actual listing, and on what timeline, is the next confirmable step.
• Whether the post-lock-up float holds its volatility down or up. Roughly half the company came free to trade on July 9. A stock that ran from HK$165 to HK$1,330 to the low HK$200s on a 5% float behaves differently once the float is ten times larger.
• The M3 Pro model. TEXXR’s corpus has MiniMax working on a 2.7-trillion-parameter successor to M3’s 428B, reportedly open-source-bound as early as the third quarter of 2026 — the first real test of whether coverage reaccelerates off its current trough.
• Status of the Anthropic distillation complaint. It has sat as a controversy edge since February 2026 with no confirmed resolution, restriction, or settlement logged since.
• The gap between the loss and the ARR claim. A $1.87B 2025 net loss against a claimed $300M in annualized recurring revenue by May 2026 is a spread TEXXR’s corpus has not yet recorded MiniMax narrowing in a confirmed edge.

For the wider pattern this name sits inside — the tiger cohort, the listing wave, the compute ceiling all of them share — see the China AI stack.

MiniMax Group lives or dies by the same constraint as these names — small equal-weighted groups, one constraint each. Not investment products: see all categories.