GLM-5.3-Flash

Specs

Input price

$0.15/M

Cached input price

$0.03/M

Output price

$0.50/M

Context window

1M tokens

Capabilities

• 320B total / 18B active MoE under a clean MIT license
• Native vision understanding from pre-training (30T-token multimodal corpus)
• Hybrid sparse-plus-linear attention with Manifold-Constrained Hyper-Connections
• reasoning_effort control (low, high, max) from flash-tier to max-effort runs

Best for

• Multimodal document pipelines: scanned records, charts, and screenshots
• High-volume agentic coding at flash-tier cost (Terminal-Bench 2.1 at 84.3)
• Long-context repo-level work on a budget

Limitations to keep in mind

• Vendor-reported benchmarks on new harnesses - treat as directional until third-party reruns
• Vision path is not a diagnostic or DICOM tool
• 1M-token context not yet verified against third-party runners

HIPAA-compliant hosting

GLM-5.3-Flash is available under our HIPAA-compliant design-partner program, with a signed Business Associate Agreement (BAA), encryption in transit and at rest, and access controls. We are currently onboarding design partners, with general availability coming soon. Run the model behind a unified OpenRouter-style API and swap to another model with a single parameter.

Pricing in context

Open source models like GLM-5.3-Flash can be served efficiently on optimized inference infrastructure, with savings passed through to you. Exact savings depend on the model and your volume, but open source inference is typically a fraction of the per-token cost of closed-source frontier models on Azure OpenAI or AWS Bedrock - without cloud egress lock-in or minimum commitments.