GLM-5.2: Open-Weight AI Model for Long-Horizon Coding

Chinese AI startup Z.ai has released GLM-5.2, a 753-billion-parameter language model for long-horizon autonomous coding. What's interesting is that Z.ai is making the model's core weights available under an open-source license. This means organizations can download, customize, and deploy GLM-5.2 on their own infrastructure. The model supports a context window of up to one million tokens. It uses a new optimization called IndexShare, which reduces computational operations by nearly three times at maximum context length. GLM-5.2 also features an upgraded Multi-Token Prediction layer to boost decoding performance, potentially increasing accepted token length by 20%. Users can choose between "Max" and "High" reasoning modes for different performance needs. The bottom line is this release offers businesses more control and flexibility in deploying advanced AI for coding and engineering tasks.