Developer Successfully Runs GLM 5.2 on Low-End Computer

Jul 9, 2026

AI Summary

A developer has managed to run the GLM 5.2 language model on a computer with limited resources, achieving communication speeds of 0.1 tokens per second. The project, named Colibrì, utilizes a mixture-of-experts approach to optimize performance on a machine with 32GB of RAM.

The developer tested GLM 5.2 on a standard computer to evaluate its performance without running out of memory (OOM).
The model was converted to int4 format, and various techniques were implemented to manage long context usage.
The Colibrì project activates approximately 40 billion parameters per token from a 744 billion parameter model, with only 11 GB changing per token.
The dense components of the model remain in RAM, while the routed experts are stored on disk and streamed as needed.
The project was developed on a 12-core laptop with 25 GB of RAM, and the developer is open to feedback and collaboration.

glmllmperformancetestingcomputing

Developer Successfully Runs GLM 5.2 on Low-End Computer

Related Stories

OpenAI says GPT 5.6 is the ‘preferred model’ for Microsoft Copilot amid breakup chatter

OpenAI introduces GPT-5.6 models with enhanced efficiency and cybersecurity features

OpenAI Releases GPT-5.6 Model with Enhanced Capabilities

OpenAI launches GPT-5.6 Sol, improving token efficiency for coding tasks