Overview
Run GLM-5.3 with high-performance inference on Crusoe. Quickly deploy Z.ai’s 753B open-weight model for agentic coding and long-horizon tasks.
Explore additional models
Frequently
asked questions
GLM-5.3 is Z.ai’s 753B-parameter open-weight model built for agentic coding, multi-step tool use, and long-horizon tasks, with a 1,048,576-token context window. It’s available for Serverless Inference and Self-Serve Deployments, running on Crusoe-operated infrastructure, so no prompts or outputs are sent to Z.ai.
GLM-5.3 is best when answer quality is the constraint, GLM-5.3 Flash is best for high-volume work where speed matters most, and GLM-5.3 MXFP4 QAT is best when you want to fine-tune GLM-5.3 with a smaller memory footprint. The MXFP4 QAT version is GLM-5.3 trained to run in 4-bit MXFP4 precision, which preserves more quality than quantizing after training. On Crusoe, GLM-5.3 is available for Serverless Inference and Self-Serve Deployments. GLM-5.3 Flash is available for Serverless Inference, Self-Serve Deployments, and Serverless Fine-Tuning. GLM-5.3 MXFP4 QAT is currently only available for Serverless Fine-Tuning.
On Serverless Inference, GLM-5.3 is billed per token, with separate rates for uncached input, cached input, and output tokens, and no minimum commitment. Self-Serve Deployments are billed per GPU hour based on the GPU you choose. Access it through an OpenAI-compatible API in Crusoe Intelligence Foundry. Current rates are shown in the pricing module above. You can also request a Tailored Deployment for bespoke inference optimization and a dedicated endpoint.
GLM-5.3 and DeepSeek V4 Pro are both frontier-scale open models tuned for agentic coding. Z.ai reports GLM-5.3 ahead of V4 Pro on DeepSWE (66.9 vs 62.7), AutomationBench (48.2 vs 43.2), and Agents’ Last Exam (28.5 vs 25.7), while V4 Pro leads on NL2Repo (61.1 vs 58.0) and Toolathlon (74.1 vs 73.0). Both share a 1,048,576-token context on Crusoe, so you can A/B them on your own agent traces.
GLM-5.3 is available for pay-as-you-go Serverless Inference, and for Self-Serve Deployments for a dedicated endpoint billed per GPU hour. To fine-tune a GLM model on your own data, choose GLM-5.2, GLM-5.3 Flash, or GLM-5.3 MXFP4 QAT, all available for Serverless Fine-Tuning. For an SLA-backed dedicated endpoint, contact us about Tailored Deployments.
MemoryAlloy technology is a proprietary memory fabric in Crusoe’s inference engine that shares the KV cache across nodes, with cache-aware routing that sends each request to the node most likely to have its context ready. For GLM-5.3, long agent traces and large codebases reused across calls skip redundant prefill, so responses start faster even at the full 1M-token context, and reused context is billed at the lower cached-input rate.



%204.png)