Baseten

Baseten has added GLM-5.3 with vision to its Model APIs, giving developers a hosted endpoint that can ingest images and turn them into working code. The launch targets teams building agentic coding tools and design-to-code workflows who want a fast path from a screenshot or mockup to a functional implementation without standing up their own inference stack. Baseten offers model inference infrastructure powering real-time inference for open-source and fine-tuned models , and this release extends that catalog with a multimodal frontier coding model on day one. GLM-5.3 is Z.AI's latest model for agentic engineering, with stronger coding, long-horizon agentic capabilities, and emergent cybersecurity analysis, all achieved through scaled post-training on the same base model as GLM-5.2. It runs on the same 744B-A40B MoE base as GLM-5.2, with every capability gain coming from scaled post-training across more diverse, realistic task environments including full codebases, documentation, testing tools, and multi-step workflows, and Terminal-Bench 3.0 moves from 4.6% to 28.3% over GLM-5.2. The vision capability shipped here is the work of Baseten's own post-training and model API teams, who layered image understanding onto the base model and tuned it for the image-to-code use case highlighted in the launch. Baseten was founded in 2019 to address the difficulty of deploying and scaling machine learning systems, with founders CEO Tuhin Srivastava, CTO Amir Haghighat, Chief Scientist Phil Howes, and Pankaj Gupta. The GLM-5.3 vision release fits the company's broader positioning around inference as the hard problem in production AI, and it gives founders and engineering teams a concrete way to test a top-tier coding model against real UI screenshots and design assets through a single API call.