Large Model

An open-weight model for locally programming agents, designed to work with Pi, OpenCode, and OpenHands for code understanding, modification, and testing.

Inception's diffusion language model platform offers fast text generation, reasoning, and tool invocation, and is suitable for customer service, search, programming, and multi-step agents.

Select multiple AI models through a unified API, and manage vendor routing, backup models, usage, and costs.

Gaode’s General-Purpose World Model Workshop enables users to quickly generate highly realistic AI digital worlds—capable of real-time interaction and stable operation for hours—using text and images. It can be deployed locally with a single RTX 5090, and its core model has been fully open-sourced.

Ant Group has launched a new-generation spatial perception model for robots that enhances 3D depth perception capabilities and improves performance in grasping, navigation, and environmental understanding.

Alibaba's open-source large-scale speech model supports zero-shot cloning in 3 seconds, multilingual capabilities, and command-based emotional control, enabling ultra-low-latency streaming synthesis at 150 ms.

AliQianwen, the first native language world model released on June 24, 2026, uses plain text to uniformly simulate seven major digital environments, allowing agents to practice in a "virtual world."

The V4 model series, developed through in-depth research, includes V4.1-Flash—which supports text-and-image understanding, long context, and tool invocation—as well as V4 Pro, which continues to provide services.

The open source streaming 3D reconstruction model of AntLingwave Technology can complete the scene 3D reconstruction and camera position estimation in real time based on a single common camera, combining the advantages of high precision, long sequence stable operation and low hardware requirements.

Lightweight large model with 100 billion parameters, focusing on high token efficiency and low latency, good at code completion, long document processing and light Agent interaction, cost-controlled, suitable for high-frequency calls and scenario-based tasks.

The world's first open-source quantum AI model series, through AI-driven quantum chip calibration and error correction, provides a high-performance tool chain for practical quantum computing and reshapes the quantum industry ecosystem.

Google DeepMind has introduced an autonomous robot AI model with powerful embodied reasoning capabilities that can efficiently accomplish tasks such as industrial instrumentation reading, complex task planning, and security risk prevention and control.

Alibaba launched the strongest domestic programming model, with intelligent body programming, multimodal reasoning and millions of contexts as the core, support for the independent disassembly and execution of complex tasks, taking into account the high performance and low threshold, known as the developers and enterprises of the “all-round programming assistant”.

Google's open source lightweight multimodal translation model supports 55 languages and image translations, with performance that exceeds larger models, taking into account both mobile and cloud deployments, and facilitating efficient globalized communication.

NVIDIA's open-source AI model series, featuring Nano, Super, and Ultra variants, is specifically designed for intelligent agent applications, delivering high efficiency and precision.

Runway's first universal world model simulates physical laws and dynamic environments through frame-by-frame pixel prediction technology. It supports robot training, digital human generation, and cross-domain simulation, redefining how AI understands and interacts with the world.

The world's first large model of mathematical reasoning in open source form to reach the gold medal level of the International Mathematical Olympiad (IMO), realizing the rigor of reasoning and the ability to solve difficult mathematical problems through a self-verification framework.

Google DeepMind's models for complex reasoning, software engineering, and multimodal long-form data analysis are available through the Gemini app, AI Studio, and APIs.

Tencent's latest release of 3D generated models, modeling accuracy increased by 3 times, geometric resolution of 1536³, support for 3.6 billion voxels of ultra-high definition modeling, and significant enhancement of detail expression.

Ali open source 80 billion parameters of the big model, 1:50 super sparse activation, millions of contexts, the cost down 90%, the performance is comparable to the hundreds of billions of models.

Tencent launched the open source raw image model, which natively supports 2K HD raw images, accurately parses complex semantics, and can efficiently generate high-quality images with Chinese and English fusion.

Alibaba's flagship large model with trillions of parameters, supporting ultra-long context, multi-language understanding and powerful inference programming capabilities, is built for complex tasks and enterprise-class applications.

ByteDance's open-source 36 billion parameter-long contextual big language model supports 512K tokens, a controlled mind budget, excels in inference, code and agent tasks, and is freely commercially available under the Apache-2.0 license.

OpenAI's models for visual understanding, computer operations, and natural language generation are well-suited for office work, data analysis, and automation tasks that require a balance between speed and cost.

DeepMind's advanced world model generates interactive, physically logical 3D virtual environments in real time from textual cues, and is widely used in gaming, education, and AGI research.

Google introduced a lightweight open source large language model , both high performance and easy to deploy , suitable for local development and multi-scenario applications .

Anthropic's Claude 5.5 series offers the Opus, Sonnet, and Haiku tiers, covering complex programming, knowledge work, and high-frequency summary extraction tasks.

A platform that connects experts with AI model development to optimize the quality and reliability of generative AI through human expertise.

Xiaomi's open-sourced 7 billion parameter inference macromodel, which outperforms models such as OpenAI o1-mini in mathematical reasoning and code competitions by a small margin.

Ema introduces a hybrid expert modeling system that dynamically combines multiple models to accomplish enterprise-class AI tasks at low cost and high accuracy.