MiniMax to Launch New Generation Large Model with 2.7 Trillion Parameters

It was reported that MiniMax is making rapid preparations for the launch of its next-generation large model. According to the information, the model's parameter scale will reach 2.7 trillion, which marks a new leap in its ability to handle complex tasks and logical reasoning.

The continuous growth of the parameters of large models intuitively reflects the industry's relentless pursuit of the intelligence level of the models. As a leading player in the field of artificial intelligence, MiniMax's recent technological layout not only demonstrates its firm commitment to the development of underlying models, but also represents an important move to establish a technological barrier by increasing the parameter scale in an increasingly competitive computing power market.

Industry analysts believe that a large model with 2.7 trillion parameters will be able to process massive data more deeply, improving the response speed and accuracy of the model in complex scenarios, thus playing a greater value in diverse applications such as dialogue interaction and intelligent analysis. Currently, as more technical details continue to emerge, the market is full of expectations for the final performance of this model.

Although the continuous expansion of the technical path brings higher R&D barriers, it also indicates that the artificial intelligence industry is accelerating towards a higher-level "intelligent" era. MiniMax's recent actions have undoubtedly injected strong new momentum into the current large model competition. The final market performance and actual application effects of its product will become the focus of attention in the industry.

Step Star's First AI Intelligent Phone Will Be Released, Leading Ahead of OpenAI

Step Star announced a new AI agent device, launching its AI terminal brand, agent system, and first AI agent phone. It becomes the first native agent phone from a major model company, with market release ahead of OpenAI's plan. Manufactured by Huaqin Technology, the two have formed a deep partnership.....

Step Astronomy to Launch the World's First AI Agent Phone from a Major Large Model Vendor

StepFun will launch a new AI terminal brand, agent system, and its first AI agent phone, becoming the first global LLM provider to deliver agent hardware. As AI models move to the edge, next-gen AI terminals are a strategic battleground; OpenAI plans a 2027 product, but StepFun is seizing the lead.....

Performance Improved by Over Two Times: NVIDIA Releases Nemotron-Labs-TwoTower Diffusion Language Model

Nvidia open-sourced Nemotron-Labs-TwinTower diffusion language model, which uses a "twin tower" architecture to overcome the serial decoding bottleneck of autoregressive models. It splits generation into two subnetworks, one kept frozen, enabling parallel text generation and higher throughput, providing an efficient solution for large-scale synthesis tasks.....

Intelligence Alternative to GPT-5? Qwen 3.6 27B Evaluation Shows Local Model Has Reached the Cutting-Edge Level

Qwen3.6 series overturns belief that local large models compromise. Tested on MacBook Max M5 128GB, Qwen3.6 27B with 8-bit GGUF quantization delivers incredible efficiency. It proves to be not only usable but a powerful general-intelligence tool without sacrifice, marking a new phase in local LLM deployment.....

Early Signs of Commercialization: Huang Zhenxin from Moonshot Explains Kimi's Differentiation Strategy

Large model industry enters deep water of deployment & cost battle. Moonshot AI's Kimi has clear commercialization. B-side head Huang Zhenxin says: insist on underlying architecture innovation, not mere engineering stacking. Kimi is high-performance model, will maintain this path despite high costs from global compute crunch.....