GLM-5.3-Flash

The MIT-licensed multimodal GLM: 320B total, 18B active, sees images and video, 1M context, built for cheap long-context serving.

In short. GLM-5.3-Flash is a cheaper, faster sibling of GLM-5.3 from Z.ai. It can look at images and video as well as read text, and its MIT license is as permissive as licenses get. It still needs a multi-GPU server to run, but much less than the full GLM-5.3.

Summary

GLM-5.3-Flash was released on 26 August 2026 and is the first natively multimodal model in Z.ai's GLM-5 series. It starts from a newly trained base model rather than a trimmed GLM-5.3, and Z.ai says it beats GLM-5.2 on benchmarks and real workloads at one tenth of the price while approaching Claude Opus 4.8 on coding and agent benchmarks (all self-reported).

It has 320 billion parameters in total and 18 billion active per token, 45 layers, and a 1-million-token context window. It mixes linear attention with sparse attention, and a component called IndexPool compresses the attention indexer to keep memory low at long lengths. Z.ai reports about 3 times less attention compute and 4.4 times less KV cache than GLM-5.3. It accepts text, images and video and writes text, and it is aimed at coding, office documents and financial research workflows with tool use.

The repository is plain MIT licensed with no revenue conditions. The checkpoint is mostly FP8. A GLM-5.3-FlashX variant is offered on Z.ai's own API for faster output; the weights published on Hugging Face are those of GLM-5.3-Flash.

Good for

  • A permissively licensed multimodal model for coding and computer-use agents
  • Long-context serving where KV cache size is the limiting factor
  • Teams that want MIT terms and a smaller footprint than GLM-5.3
  • Reading screenshots and interface states in agent loops

Watch out for

  • 320 billion parameters is still far beyond a single consumer GPU
  • Inference support is newer than for older architectures, though Z.ai lists SGLang, vLLM and TokenSpeed
  • The comparison with closed models is Z.ai's own
  • It outputs text only

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the GLM-5.3-Flash variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
GLM-5.3-Flash FP8 320B 18B active 736 GB 368 GB 221 GB Big-memory desktop

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.

More about MIT

See also

Frequently asked questions

What is GLM-5.3-Flash?
GLM-5.3-Flash is a cheaper, faster sibling of GLM-5.3 from Z.ai. It can look at images and video as well as read text, and its MIT license is as permissive as licenses get. It still needs a multi-GPU server to run, but much less than the full GLM-5.3.
Can I use GLM-5.3-Flash commercially?
Yes. GLM-5.3-Flash is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is GLM-5.3-Flash open source?
GLM-5.3-Flash is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run GLM-5.3-Flash on my own computer?
Only on a machine with a lot of memory. The smallest variant, GLM-5.3-Flash, needs about 221 GB at 4-bit precision. Desktops with 128 to 512 GB of unified memory can hold it, at modest speed.
When was GLM-5.3-Flash released?
GLM-5.3-Flash was released by Z.ai (Zhipu AI) on August 26, 2026.
How long a context window does GLM-5.3-Flash have?
GLM-5.3-Flash supports a context window of up to 1M tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
GLM-5.3 Z.ai (Zhipu AI) 753B 40B active 1M GLM-5.3 License (Conditional) 520 GB GPU server
MiniMax M3.1 Flash MiniMax 428B 23B active 1M MiniMax Community (Conditional) 277 GB Big-memory desktop
Qwen3.8-Flash-Next Alibaba Qwen 180B 6B active 256K Qwen Community 1.0 (Conditional) 124 GB Big-memory desktop
AREX-2 BAAI 27.4B 256K Apache 2.0 (Permissive) 19 GB Gaming GPU