LLM
GA
Open source
GLM-5.3-Flash
Cheap open-weight agentic coding with image/video input
Release2026-08-26
Context1.05M
Max output131,072 tokens
Parameters320B
Local / VRAMLocal
LicenseOpen source
Overview
GLM-5.3-Flash is a model in the LLM category from Z.ai (Zhipu AI). It was released in 2026-08-26. The vendor is based in China. Best for: Cheap open-weight agentic coding with image/video input.
Notes: Previewed as the stealth model Ox Alpha. First natively multimodal GLM-5 model.
Hardware: About 306 GiB of weights; 8-GPU node class (FP8). Some hosts serve a 256K context.
Strengths
- MIT license with open weights
- Natively multimodal incl. video input
- About one tenth of GLM-5.2 price (vendor)
Limits
- Needs an 8-GPU node to self-host
Specifications
| Vendor | Z.ai (Zhipu AI) |
|---|---|
| Category | LLM |
| Release | 2026-08-26 |
| Status | GA |
| License | Open source |
| License name | MIT |
| Commercial use | Allowed |
| Parameters | 320B |
| Active parameters | 18B |
| Architecture | MoE with hybrid linear + sparse attention |
| Context window | 1.05M |
| Max output tokens | 131,072 |
| Input modalities | Text, Image, Video |
| Output modalities | Text |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| API access | Yes |
| Pricing | $0.15 / $0.50 per 1M token (cached input ~$0.03) |
| Confidence | High |
| Vendor country | China |
| Vendor type | Open lab |
Pricing & hardware
$0.15 / $0.50 per 1M token (cached input ~$0.03)
This model can be self-hosted or run locally. API access is available.