Overview
Run GLM-5.3 Flash on Crusoe. Customize and deploy Z.ai’s fast 320B open-weight model with image input and a 1M-token context, built for coding and agents.
Explore additional models
Frequently
asked questions
GLM-5.3 Flash is Z.ai’s fast, cost-efficient model in the GLM-5.3 family, a mixture-of-experts model with 320B total parameters and about 18B active per token. It is the first natively multimodal model in the GLM-5 series, accepting text and image input, and supports a 1,048,576-token context. On Crusoe, it is available for Serverless Inference, Self-Serve Deployments, and Serverless Fine-Tuning. You can also request a Tailored Deployment for bespoke inference optimization and a dedicated endpoint.
GLM-5.3 is the larger 753B model built for maximum reasoning depth, while GLM-5.3 Flash is 320B and tuned for speed, with fewer active parameters per token and native image input. Both support a 1,048,576-token context, so you can route between them without re-chunking inputs. Choose Flash for latency-sensitive and high-volume work, and GLM-5.3 when output quality matters more than response time.
For Serverless Inference, GLM-5.3 Flash 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.
Z.ai reports GLM-5.3 Flash outperforming the previous-generation GLM-5.2 across benchmarks at a fraction of the price, and approaching leading closed models on coding and agentic tasks. With only about 18B parameters active per token, it delivers that quality at much higher throughput than the full GLM-5.3.
GLM-5.3 Flash is available for Serverless Inference, Self-Serve Deployments, and Serverless Fine-Tuning. To fine-tune, upload a JSONL or Parquet dataset and run a LoRA job with pre-configured defaults, then deploy the checkpoint in one click to a dedicated endpoint through Self-Serve Deployments, billed per GPU hour, or download the weights. For an SLA-backed dedicated endpoint, contact us about Tailored Deployments.
Yes, GLM-5.3 Flash is the first natively multimodal model in Z.ai’s GLM-5 series, accepting both text and image input and returning text. That makes it useful for screenshot-to-code, document and chart understanding, and agents that need to read what’s on screen. The full-size GLM-5.3 is text-only, so Flash is the GLM-5.3 model to choose when inputs include images.
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 Flash, high-volume agent loops and repeated long prompts skip redundant prefill, so responses start faster at scale, and reused context is billed at the lower cached-input rate.



%204.png)