Hugging Face Releases GLM-5.2 With Million-Token Context

Tech · 2026-08-18 23:20:16

Verified

Hugging Face released the GLM-5.2 model on June 17, 2026. The company designed the release for long-horizon workloads. The architecture supports a one-million-token context window. Training data was expanded to cover programming-agent scenarios.

The DeepSeek-V4 model also provides a million-token context capacity.

Sources

• Hugging Face Blog: GLM-5.2: Built for Long-Horizon Tasks
• Hugging Face Blog: DeepSeek-V4: a million-token context that agents can actually use
• Hugging Face Blog: GGML and llama.cpp join HF to ensure the long-term progress of Local AI
• Hacker News: Launch HN: machine0 (YC S26) – Persistent CPU and GPU VMs from the CLI
• Hugging Face Blog: Introducing HELMET: Holistically Evaluating Long-context Language Models
• Hugging Face Blog: Cosmopedia: how to create large-scale synthetic data for pre-training Large Language Models