GLM-5.3-Flash — reviews, specs & pricing

320B-A18B hybrid-attention MoE (MIT), first natively multimodal GLM-5 model; ran anonymously on OpenRouter as 'Ox Alpha' before launch.

Summary

GLM-5.3-Flash is a 320B-parameter hybrid-attention Mixture-of-Experts (MoE) model from Zhipu AI, activating just 18B parameters per token for ultra-fast generation. Originally evaluated under the anonymous codename 'Ox Alpha' on OpenRouter, it serves as a lightweight, high-throughput flagship for chat and coding tasks under a permissive MIT license. It bridges the gap between massive frontier model reasoning and practical, low-latency deployment.

Sample use case

Deploying scalable, enterprise-grade developer assistants and multi-tenant code review bots that require low-latency code completion, refactoring, and technical explanations across large codebases without incurring prohibitive commercial API costs.

Specifications

  • Provider: zhipu ai
  • License: open
  • Parameters: 320B
  • Context: 1049k tokens
  • Input price: $0.15/M tok
  • Output price: $0.5/M tok
  • Released: 2026-08-26

Pros

  • Permissive MIT open-weights license
  • High throughput with only 18B active parameters
  • Strong conversational and code generation performance
  • Efficient hybrid-attention architecture

Cons

  • Substantial total VRAM required to host all 320B weights
  • Complex deployment and orchestration requirements
  • Lower peak reasoning depth than heavy frontier models

Average rating 0.0 from 0 community reviews on Reviuws.

Frequently asked questions

What is GLM-5.3-Flash?

GLM-5.3-Flash is a 320B-parameter hybrid-attention Mixture-of-Experts (MoE) model from Zhipu AI, activating just 18B parameters per token for ultra-fast generation. Originally evaluated under the anonymous codename 'Ox Alpha' on OpenRouter, it serves as a lightweight, high-throughput flagship for chat and coding tasks under a permissive MIT license. It bridges the gap between massive frontier model reasoning and practical, low-latency deployment.

How much does GLM-5.3-Flash cost?

GLM-5.3-Flash costs $0.15 per million input tokens and $0.5 per million output tokens.

Is GLM-5.3-Flash open source?

GLM-5.3-Flash is released under the open license.