MiniMax Joins Stock Connect as Shares Jump More Than 17%: H3 and Market Impact Explained

Introduction
MiniMax Group joined the eligible securities lists for Southbound trading through both Shanghai–Hong Kong Stock Connect and Shenzhen–Hong Kong Stock Connect on August 6, 2026.
The company trades in Hong Kong under the stock code 00100 and the short name MINIMAX-W.
The adjustment allowed eligible mainland investors to buy and sell MiniMax shares through the Stock Connect framework from the effective date.
The market reacted quickly. At the time the original AIBase report was published at 10:41 a.m. on August 6, MiniMax shares had risen by more than 17%.
The inclusion arrived during a period of heightened attention around MiniMax H3, the company’s recently open-sourced general-purpose multimodal video model.
MiniMax presents H3 as a system that can understand text, images, video, and audio together, then generate videos with native stereo sound at resolutions up to 2K and durations of up to 15 seconds.

The combination of a broader investor-access channel, a rising share price, and a high-profile open model helped draw additional attention to the company.
However, the events should be separated carefully:

• Stock Connect inclusion expands access for eligible investors.
• The intraday share-price increase reflects market trading on that day.
• H3’s technical performance is evaluated through changing leaderboards and product tests.
• Analyst price targets represent research opinions, not guarantees. MiniMax Was Added to Both Southbound Stock Connect Lists On August 5 and August 6, the Shanghai and Shenzhen exchanges announced adjustments to the securities eligible for Southbound trading. The additions included:
• MiniMax-W — 00100
• Luxshare Precision — 02475
• Chaozhou Three-Circle — 06951 The Shenzhen adjustment took effect on August 6. The Shanghai notice stated that the changes would take effect from the next Southbound Stock Connect trading day. For MiniMax, the practical result was that eligible mainland investors could trade the company’s Hong Kong-listed shares through the relevant Stock Connect channels. What Is Southbound Stock Connect? Stock Connect is a mutual-market-access framework linking mainland Chinese exchanges with the Hong Kong market. The Southbound direction allows eligible mainland investors to trade specified Hong Kong-listed securities through mainland brokers and exchange infrastructure. The Northbound direction allows eligible Hong Kong and international investors to trade designated mainland-listed securities. A simplified structure looks like this: Direction Investor Location Market Accessed Southbound Mainland China Eligible Hong Kong-listed securities Northbound Hong Kong and overseas Eligible Shanghai- or Shenzhen-listed securities MiniMax’s inclusion relates to Southbound trading. Inclusion Does Not Mean Every Mainland Investor Can Trade the Stock Stock Connect access remains subject to:
• Investor eligibility rules
• Broker support
• Trading calendars
• Daily quota and market arrangements
• Currency conversion
• Applicable fees and taxes
• Stock Connect trading rules
• Securities remaining on the eligible list The list can also change over time. Inclusion makes the stock available through the channel. It does not guarantee that capital will flow into the stock or that its price will rise. Why Stock Connect Inclusion Matters A Hong Kong-listed company can already be traded by investors with direct access to the Hong Kong market. Southbound inclusion adds a more convenient route for eligible mainland capital. This can matter in several ways. A Larger Potential Investor Base Some mainland investors prefer or need to trade Hong Kong securities through Stock Connect rather than opening and operating a separate Hong Kong brokerage account. Eligibility can therefore expose a company to a broader pool of potential buyers. Greater Visibility Stocks added to the channel may receive more attention from:
• Mainland institutions
• Mutual funds
• Securities analysts
• Financial media
• Quantitative strategies
• Retail investors eligible for Southbound trading That visibility can improve trading activity, although the actual outcome varies by stock. Possible Liquidity Effects More participants can contribute to higher turnover and deeper liquidity. The result is not automatic. Liquidity still depends on:
• Investor demand
• Free float
• Valuation
• News flow
• Business performance
• Broader market conditions Future Southbound Ownership Can Be Tracked The Shanghai Stock Exchange publishes information on the aggregate quantity of eligible Hong Kong securities held through Stock Connect. That allows market participants to observe Southbound holdings after inclusion. The data should not be interpreted as a simple buy or sell signal. It is one input among many. MiniMax Shares Rose More Than 17% After the Adjustment The AIBase report stated that MiniMax’s share price had increased by more than 17% by publication time on August 6. Other market-data pages showed the stock trading sharply higher during the session, with the exact percentage changing throughout the day. The most accurate way to describe the event is: MiniMax shares rose more than 17% during the morning session on August 6, 2026, after the Stock Connect adjustment became effective. It is difficult to attribute a share-price movement to one cause with certainty. Possible factors included:
• New Southbound eligibility
• Expectations of mainland investor demand
• Recent attention around MiniMax H3
• Analyst commentary
• Short-term trading momentum
• Wider sentiment toward AI stocks
• Company-specific financing and product news Stock prices can reverse quickly. An intraday rise is not evidence of a permanent valuation change. MiniMax Was Already a Public Company Before the Stock Connect Change MiniMax completed its Hong Kong initial public offering in January 2026. Its listing documents identified MiniMax Group Inc. as a Cayman Islands-incorporated company controlled through weighted voting rights. The company’s shares began trading on the Hong Kong Stock Exchange on January 9, 2026. The final IPO price was HK$165 per share, according to the offering results reported at the time. The Stock Connect event did not represent a new listing. It represented a change in the channels through which eligible mainland investors could access the already-listed shares. Stock Code and Corporate Identity Item Detail Listed company MiniMax Group Inc. Hong Kong stock code 00100 Trading short name MINIMAX-W Listing date January 9, 2026 Listing market Hong Kong Stock Exchange Main Board Structure Weighted voting rights company The “-W” suffix identifies a company with a weighted voting rights structure under Hong Kong market conventions. MiniMax H3 Has Become a Major Technology Catalyst The source article links market attention to the release of MiniMax H3. MiniMax officially open-sourced H3 on August 3, 2026. The model is described as a general-purpose, omni-modal generative system. It supports context containing:
• Text
• Images
• Video
• Audio It can generate synchronized video and stereo audio. Official H3 Output Specifications MiniMax’s public model card lists the following specifications: Feature Specification Output duration 4–15 seconds Default visual scale Shorter side at 768 pixels High-resolution workflow Up to 2K through H3-Regenerate-2K Frame rate 24 FPS Audio 32 kHz stereo Common aspect ratios 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 and others Stable dialogue languages 11 listed languages, with partial support for others The model supports several task modes. Text and Keyframe Generation The H3-Base-FL2VA checkpoint supports:
• Text-to-audio-video
• First-frame-to-audio-video
• Last-frame-to-audio-video
• First-and-last-frame-to-audio-video The workflow can use zero, one, or two input images. Multimodal Reference Generation The H3-Base-Ref2VA checkpoint supports reference-based generation using:
• Up to 9 images
• Up to 3 video clips
• Up to 3 audio clips
• Up to 12 files in total Reference video and audio durations are subject to limits documented in the model card. The model can therefore use more than one type of source material in the same generation workflow. H3’s Open Release Includes Important Components—but Not the Entire Hosted System MiniMax divides the full H3 system into three main parts:
• H3-Context-IR
• H3-Base
• H3-Regenerate-2K H3-Context-IR H3-Context-IR interprets free-form multimodal inputs and converts them into a structured intermediate representation. MiniMax says it handles tasks such as:
• Instruction parsing
• Cross-modal association
• Temporal understanding
• Logical relationships among references
• Filling in underspecified details where appropriate The full hosted Context-IR orchestration system is not included in the open-source release. MiniMax provides an API and prompting guidance instead. That distinction is important when comparing local deployment with the official hosted experience. H3-Base H3-Base generates synchronized video and audio, with the open workflow producing 768p results. Two task-specific checkpoints are available:
• FL2VA
• Ref2VA The released checkpoints use BF16 precision and include the model components needed for their supported task family. H3-Regenerate-2K The high-resolution stage takes the 768p result and the original context, then regenerates the output at 2K. This is not a simple conventional upscale. MiniMax describes it as a context-aware regeneration process intended to improve details and visual fidelity. H3 Uses a Unified Audio-Video Architecture MiniMax’s public repository explains that H3 encodes the different input modalities and organizes them into a unified packed sequence. The architecture uses:
• H3-Encoder
• H3-VisualVAE
• H3-AudioVAE
• H3-Omni-Transformer The H3-Omni-Transformer predicts video and audio latent representations together. Those latents are then decoded into:
• Video
• Stereo audio The joint-generation design is intended to improve synchronization between what happens visually and what is heard. That can be relevant for:
• Dialogue
• Sound effects
• Environmental sound
• Music-driven scenes
• Reference-audio workflows H3’s Leaderboard Position Needs Precise Wording The original AIBase article says H3 ranked first globally on both an Artificial Analysis video-editing leaderboard and an Arena image-to-video leaderboard. Leaderboard positions are dynamic, and the available snapshots reviewed for this publication do not support that statement as a universal current ranking. Artificial Analysis Image-to-Video Results On the Artificial Analysis image-to-video leaderboard snapshot reviewed on August 6: Without audio:
• Gemini Omni Flash
• MiniMax H3
• Dreamina Seedance 2.0 720p With audio:
• Dreamina Seedance 2.0 720p
• Gemini Omni Flash
• MiniMax H3 H3 did lead the open-weight category in both versions of the image-to-video leaderboard. That is a meaningful distinction. A precise description is therefore: MiniMax H3 ranked among the leading image-to-video models and was the highest-ranked open-weight model in the Artificial Analysis snapshots reviewed on August 6. Arena Image-to-Video Position The Arena image-to-video page dated August 2 listed H3 among the top models, but not as the overall number-one model in the available snapshot. Arena rankings can change as new votes arrive. They can also vary based on:
• Model version
• Audio inclusion
• Resolution
• Task type
• Leaderboard date
• Statistical confidence
• User preference distribution For publication, the leaderboard date and category should always be included. Video Editing Is a Different Task Image-to-video and video editing are not identical evaluations. A model can perform strongly in one and less strongly in another. The current Artificial Analysis video-editing leaderboard reviewed for this article was led by other models and did not show H3 as the overall number-one entry in the default current-model view. The broad “global first on both lists” description should therefore be treated as a source-era or filter-specific claim rather than a stable fact. Hugging Face Interest Rose After the Open Release MiniMax released H3 weights through its official Hugging Face organization and GitHub repository. The Hugging Face page exposes:
• Model files
• Task-specific checkpoints
• Diffusers integration
• Deployment instructions
• Local serving guidance
• Prompting documentation
• Community discussions
• Example Spaces The model quickly attracted community attention and spawned multiple demonstration Spaces. The AIBase article says H3’s Hugging Face popularity surpassed DeepSeek V4 Flash and made it one of the community’s most closely watched models. Hugging Face “trending” positions are highly dynamic. They can change based on:
• Recent likes
• Downloads
• Updates
• Community activity
• Search filters
• Time window
• New model releases A trending rank is useful as a measure of short-term attention. It is not a stable benchmark of model quality. Developers Can Run H3 Locally, but the Hardware Requirement Is Significant The official repository provides local-serving examples for SGLang. A documented FL2VA command is: sglang serve \ --model-path MiniMaxAI/MiniMax-H3 \ --num-gpus 4 \ --ulysses-degree 4 \ --performance-mode speed \ --host 0.0.0.0 \ --port 30010 \ --model-variant fl2va For Ref2VA: sglang serve \ --model-path MiniMaxAI/MiniMax-H3 \ --num-gpus 4 \ --ulysses-degree 4 \ --performance-mode speed \ --host 0.0.0.0 \ --port 30011 \ --model-variant ref2va These examples use four GPUs. The real hardware requirement depends on:
• Checkpoint
• Precision
• Resolution
• Duration
• Parallelism
• Inference framework
• Memory optimization
• Performance target Open weights do not mean inexpensive local deployment. Video generation remains computationally demanding. Industry Figures Publicly Noticed H3 The source article says several prominent technology and investment figures publicly commented on or followed the H3 release. It names:
• Emad Mostaque, founder of Stability AI
• Justine Moore, partner at Andreessen Horowitz Public attention from well-known industry figures can help a model gain visibility. It should not be treated as an independent technical certification. A social post may reflect:
• Interest in the architecture
• Interest in open weights
• Curiosity about results
• Support for the open ecosystem
• A general view of market direction It is not the same as a controlled benchmark, code audit, or investment recommendation. Jefferies Maintained a Buy Rating and HK$1,118 Target The AIBase article also points to an analyst report from Jefferies. Jefferies initiated coverage of MiniMax in February 2026 with a Buy rating and a price target of HK$1,118. More recent syndicated reporting said the broker reiterated that rating after the H3 launch. The target is substantially above the stock’s trading price around the Stock Connect inclusion date. Analyst targets are based on assumptions. Those assumptions may cover:
• Future revenue growth
• User adoption
• Video-generation demand
• API usage
• Enterprise customers
• Model cost
• Gross margins
• International expansion
• Capital spending
• Long-term profitability
• Valuation multiples If those assumptions change, the target can change. Analyst Ratings Are Not Guarantees A price target does not guarantee that a stock will reach that level. Analysts may:
• Revise models
• Change forecasts
• Update ratings
• Adjust discount rates
• Respond to financing or dilution
• Reassess competition
• Change the target after earnings Readers should treat the HK$1,118 figure as Jefferies’ research opinion, not as an established future value. Cost Efficiency Is Central to the H3 Investment Argument The source says Jefferies viewed H3’s cost advantage as a factor that could support enterprise adoption and commercialization. Cost is especially important in AI video because generation consumes substantially more compute than ordinary text inference. Commercial customers may evaluate:
• Cost per generated second
• Success rate
• Number of retries
• Reference consistency
• Audio quality
• Resolution
• Generation time
• Rights and safety controls
• API reliability
• Local-deployment options A lower list price does not always mean a lower production cost. A cheap model that requires many retries may cost more per accepted clip than a more expensive model with a higher first-pass success rate. The useful metric is: Total generation cost ÷ Number of usable final outputs Open weights add another option. Some organizations may choose to operate the model on their own infrastructure, especially when they need:
• Data control
• Custom deployment
• Integration with internal pipelines
• Model experimentation
• Hardware-specific optimization Self-hosting also adds infrastructure, engineering, and operational costs. H3 Strengthens MiniMax’s Position in Multimodal AI MiniMax has developed products across several media types. Its public product and developer ecosystem includes:
• Text models
• Coding and agent tools
• Speech
• Music
• Images
• Video
• Consumer applications
• API services H3 is important because it combines several modalities in one video-generation system. The model can use:
• Text to define the task
• Images to define subjects or frames
• Video to define motion or visual context
• Audio to define voice, rhythm, or sound context This makes it more than a conventional text-to-video model. Its commercial relevance depends on whether customers can obtain repeatable results in workflows such as:
• Advertising
• E-commerce
• Entertainment
• Game content
• Social video
• Virtual characters
• Product visualization
• Video localization
• Creative previsualization The open release also creates a developer ecosystem outside MiniMax’s hosted application. Stock Connect Does Not Remove the Company’s Business Risks The source article emphasizes the positive attention around MiniMax. A balanced publication should also distinguish access and sentiment from business fundamentals. MiniMax remains exposed to risks common to frontier AI companies. High Compute Spending Training and serving multimodal models requires large infrastructure investment. Video models are particularly expensive to operate. Fast Product Cycles Competitors can release new models quickly. Leaderboard positions may change within days or weeks. Monetization Uncertainty Strong user growth does not automatically produce sustainable margins. Model Commoditization Open-source and hosted competitors can put pressure on pricing. Content and Regulatory Risk Video-generation systems face issues involving:
• Copyright
• Likeness rights
• Harmful content
• Misleading media
• Regional regulation
• Platform moderation Financing and Dilution MiniMax has continued to raise capital for AI infrastructure and research. New shares, placements, or convertible securities can dilute existing shareholders. Valuation Volatility AI stocks can trade on expectations far ahead of current financial performance. That can create sharp upward and downward movements. What the August 6 Event Actually Confirms The following points are supported by the available exchange, company, and market records:
• MiniMax is listed in Hong Kong under stock code 00100.
• Its shares began trading on January 9, 2026.
• MiniMax was added to the eligible lists for Southbound Stock Connect in August 2026.
• The Shenzhen adjustment took effect on August 6.
• The stock rose by more than 17% during the morning session reported by AIBase.
• MiniMax H3 was open-sourced on August 3.
• H3 supports multimodal reference inputs and synchronized audio-video generation.
• H3 ranked among the leading image-to-video models in current public evaluations.
• H3 led the open-weight image-to-video categories in the Artificial Analysis snapshot reviewed.
• Jefferies published a Buy rating and HK$1,118 target. What Should Not Be Inferred Automatically The following conclusions are not established by the event:
• Stock Connect inclusion guarantees continuing share-price gains.
• Southbound investors will necessarily buy the stock.
• H3 is number one on every video benchmark.
• Hugging Face popularity proves commercial success.
• An analyst target guarantees a future market price.
• Open weights guarantee low-cost self-hosting.
• Technical performance automatically produces profitability.
• One strong trading session establishes a long-term trend. A Short Timeline Date Event December 31, 2025 MiniMax published its Hong Kong global-offering prospectus January 9, 2026 MiniMax shares began trading in Hong Kong February 19, 2026 Jefferies initiated coverage with a Buy rating and HK$1,118 target August 3, 2026 MiniMax officially open-sourced H3 August 5, 2026 Shanghai exchange announced MiniMax’s addition to Southbound eligibility August 6, 2026 Shenzhen adjustment took effect; MiniMax shares rose sharply during the morning session