MiniMax has released the weights of the open-source H3-Base model on the HuggingFace platform. This model is a dense, single-stream Omni-Transformer with 33 billion parameters, equipped with a VAE featuring 16x spatial and 4x temporal compression. Furthermore, its usage cost via API is one-third of the price of Seedance 2.0.
The H3 model is a universal system for generating all modalities, unlike models designed only for text-to-video or image-to-video conversion. It is capable of understanding contexts from text, images, video, and audio, and then generating video with native stereo sound. Architecturally, H3 does not consist of a sequential connection of video and audio models.
The main network, H3-Omni-Transformer, is implemented as a dense, single-stream Transformer with 33 billion parameters, with about 13 billion parameters located in the AdaLN branch, whose modulation results can be pre-calculated and saved for lighter inference. Text encoding is handled by H3-Encoder, visual data by H3-Encoder and H3-VisualVAE, and audio by H3-AudioVAE; all these elements are integrated into a single sequence that is jointly predicted for video and audio latents.
MiniMax improved H3-VisualVAE by implementing 16x spatial compression and 4x temporal compression. After additional patching, the effective spatial downsampling of video tokens entering the Transformer reaches a 32x level, which reduces computational costs during high-resolution video generation. The complete MiniMax H3 suite includes three components: H3-Context-IR, H3-Base, and H3-Regenerate-2K. H3-Context-IR interprets user-provided text, images, video, and audio, structuring complex multimodal instructions into intermediate representations suitable for generative models. H3-Base is responsible for the actual generation of video and audio with primary output at 768P resolution. The H3-Regenerate-2K component uses context regeneration to obtain a 2K version from the 768P result while preserving the original context. Currently, the H3-Base component has open weights.
The pricing policy makes H3 a significant market disruptor: 2K resolution generation costs 0.8 yuan per second, and 768P resolution is cheaper, with free audio input and up to five free images. In terms of performance, H3 demonstrates competitiveness against ByteDance's Seedance 2.0 at one-third the cost, and local deployment on two RTX 5090 cards or one RTX 6000 reduces costs to zero. With its flagship capabilities, favorable pricing, and open weights, the initiative for video creation becomes accessible to every developer.
The release of this model intensifies competition in the AI video generation market. Seedance 2.0 previously dominated due to unified joint audio and video generation, and MiniMax H3 directly challenges this position by offering comparable functionality at significantly lower costs. For content creators and developers, H3 provides flexibility unavailable on closed platforms, including the option for local deployment for data-sensitive workflows, custom fine-tuning, and integration with existing production pipelines. The open-weights strategy aligns with the global trend where Chinese AI companies use open source and aggressive pricing to counter closed ecosystems. Thus, H3 is reshaping the economics of AI video generation, challenging the assumption that advanced video models must be expensive cloud services, and potentially accelerating adoption in advertising, e-commerce, film production, and content creation, where usage was previously limited by cost sensitivity.
XPower Technology has unveiled the RPP AE7100E architecture chips at the WAIC event. These nail-sized microchips are capable of supporting models with parameters ranging from 400 billion to 1.6 trillion by utilizing distributed computing, while maintaining minimal token cost.
XPower Technology has established itself as a significant player in the domestic AI computing market at WAIC 2026. The company demonstrated its matrix of artificial intelligence accelerators based on its proprietary Reconfigurable Parallel Processor (RPP) architecture. This architecture covers cloud, edge, and end computing scenarios. The company's core idea directly challenges the industry's prevailing doctrine regarding the necessity of using giant chips.
The AE7100E chips, which are about the size of a fingernail, can be integrated in quantities of 12 per accelerator card. This configuration allows for running large language models with 400 billion to 1.6 trillion parameters through distributed computing, achieving an industry-optimal cost efficiency per token. XPower CEO, Li Yuan, noted that the shift from training-oriented computing to inference-oriented computing is a key trend. He emphasized that AI programming has become a critical use case for inference, as the company's experience has shown that AI can now perform CPU compiler development, which previously required a professional team for years, and this capability can be scaled for rapid application across various tasks.
Since 55% of model requests in global data centers are directed towards open-source models, the inference market is accessible not only to a few giants but to any enterprise, creating huge demand for cost-effective inference infrastructure. A key observation is the inefficiency of large chips during inference: decoding requires a computation-to-data ratio of 1:1, whereas GPUs provide a 1000:1 ratio, leading to throughput loss. XPower's RPP architecture uses distributed small chips, enabling fine specialization within a single server. Furthermore, XPower's approach replaces expensive HBM technology with the more mature LPDDR, asserting that three elements are fundamentally necessary for AI inference: high capacity, sufficient bandwidth, and low cost. LPDDR technology meets all these requirements at a significantly lower cost than HBM. The configuration of 12 chips per card allows for detailed memory partitioning and parallel data processing between chips within a single server form factor. This distributed server architecture differentiates XPower from the NVIDIA Groq approach, which also uses distributed multi-chip designs, but on a much larger cluster scale. The inherent flexibility of the RPP architecture, balancing general and domain-specific computations, allows one chip to be used for deployment scenarios in the cloud, at the edge, and on the end device.
During WAIC, the company showcased its product line, ranging from the AE7100E chip to high-density server racks with 8 cards. The company's transition from edge computing to cloud is a strategic bet that the inference market—which XPower estimates to be ten times larger than the training market by total addressable value—will be captured not by the largest computational power of a single chip, but by the most cost-effective distributed systems. Although confirmation of this hypothesis will depend on real-world implementation results, the RPP architecture represents one of the most original domestic approaches to computing, challenging the consensus on large chips by asserting that the future of AI inference belongs not to single monolithic processors, but to clusters of coordinated small chips.
