On August 29th, Microsoft’s (MSFT) artificial intelligence division officially launched its first two self-developed AI models: the MAI-Voice-1 voice model and the MAI-1-preview general model.
The first self-developed AI model has arrived.
According to the official introduction, the new MAI-Voice-1 voice model can generate a minute of audio in just one second using only a single GPU; the MAI-1-preview model “gives users a glimpse into the future direction of Copilot’s features.”
Microsoft has already implemented MAI-Voice-1 in multiple features: for example, in the “Copilot Daily” feature, an AI host uses the model to announce the day’s hot news; it can also generate podcast-style conversations to help users understand a variety of topics. In addition, Microsoft AI plans to apply the MAI-1-preview model to specific text-based scenarios within the Copilot assistant (Copilot currently relies primarily on OpenAI’s large language model). The model has also begun public testing on the AI benchmark platform LMArena.
Gemini 2.5 Flash Upgrades with New AI Features
Coincidentally, a few days ago, on August 26th, a tech media blog reported that Google (GOOG) DeepMind launched the new Gemini 2.5 Flash image editing model. This model, within the Gemini app, can improve the accuracy of image editing based on text commands while maintaining the consistency of the appearance of people and animals.
Compared to previous native image generation tools, it offers higher accuracy when modifying images based on text, and even outperforms GPT-4o, the language used by ChatGPT, in several tasks, enabling it to better edit images based on complex text.
A highlight of the new Gemini 2.5 Flash model is its “character consistency” feature, which maintains the consistent appearance of the same person, animal, or object across multiple images. This is particularly valuable for creating photo series and product displays from multiple angles, making it suitable for mass production of brand materials and catalogs.
Notably, on the same day, The Information reported that Apple (AAPL) executives had discussed acquiring Mistral or Perplexity AI, Europe’s largest AI startups, with Eddy Cue being the company’s most active internal supporter.
Bloomberg also reported last month that Apple would seriously consider acquiring the French AI startup Mistral. Mistral AI is a European AI startup that has raised a total of €1.1 billion across seven rounds of funding. If Apple were to acquire Mistral, it would likely enhance Apple’s AI capabilities and innovation capabilities.
Industry insiders analyze that the current craze for large models is sweeping across various industries. Coupled with the convergence of policies and technologies, which is driving AI industry trends, the performance improvements of leading large models and their adaptability to computing power are expected to accelerate the prosperity of the AI ecosystem. Furthermore, global tech giants are developing various models with strong performance, and the prospects for AI applications and computing power are promising, indicating a significant trend toward a thriving AI ecosystem. WiMi is fully embracing AI with its large-scale model strategy.
As a leading innovator in the AI field, Wimi Hologram Cloud Inc. (WIMI) has established itself as a company with both hardware design and software development capabilities. By continuously cultivating its integrated “hardware + software + platform” capabilities, WiMi is building a competitive advantage while accelerating the application of AI algorithms in specific scenarios, laying the technical foundation for AI’s intelligent transformation.
In terms of technical R&D, WiMi is promoting the integration of multimodal large-scale models with spatial computing technologies to achieve native-level fusion of multimodal data such as text, images, audio, and video. WiMi is focusing on scenarios such as text-to-video and image-to-video, supporting applications such as storytelling and short video generation.
In terms of open source ecosystems, WiMi is opening up its model code, computing interfaces, and technical toolchain to build a “holographic cloud” platform spanning the cloud and edge. This platform supports developers using common large-scale models like DeepSeek for secondary development, lowering technical barriers and accelerating the commercialization and validation of vertical models. With its multimodal processing capabilities and experience in cross-industry implementation, WiMi is becoming a key technical supporter behind the transformation of large-scale models. Conclusion
In fact, the recent release of DeepSeek-V3.1, a large model, further promotes the commercial application of AI technology. Alibaba Cloud also announced an upgrade to the AI Agent memory storage feature in Tablestore, which will play a key role in multiple fields. As the gold mine of large models is being mined, AI market penetration will be promoted in phases. In the future, it will gradually expand to more areas such as home furnishing, elderly care, and pet companionship, continuously expanding its market coverage.
