GLM-5.3 is the strongest open-weights coding model

Large language models (LLMs) are advanced AI systems trained on massive text datasets to understand, generate, and reason about human language, powering applications such as chatbots, code generation, summarization, and scientific discovery.

Systems trained on data to learn patterns and make predictions or generate outputs, such as text, images, or decisions, by approximating relationships between inputs and outputs without being explicitly programmed for each task.

Coverage of China's AI models, chip access, and competition with US labs.

AI models whose trained weights are published for anyone to download, run, and modify.

The practice of measuring and comparing the capabilities of AI models through standardized benchmarks and evaluation protocols.

Writing, reviewing, refactoring, and shipping code.

Tools for building and managing AI systems. Supports development, testing, and deployment.

Models that process and generate language. Used in communication and content tasks.

Automatically producing source code from natural language instructions or other code, used in developer tools and AI assistants.

Z.ai's 320B-A18B open-weight model, released under the MIT License and previewed anonymously as Ox Alpha.