GLM-5.2 launch and GLM Coding Plan

What's new

• GLM-5.2 model: open-weight coding-focused model, 1M token context.
• MIT license: weights released under a permissive license — no commercial restrictions.
• GLM Coding Plan: subscription with unlimited access at ~1/10 the price of top Anthropic tiers.
• API: launched alongside the release, available via z.ai and partners.
• Benchmark performance: competitive with Claude Opus 4.8 and GPT-5.5 on coding tasks.

Why it matters

This is the third strong open-weight coding model in a few months (after DeepSeek V3.5 and Qwen 3 Coder). The 10× price advantage is significant — for high-volume teams (5+ developers, heavy agent runs) it can be a rational replacement for Anthropic or OpenAI. Vercel CEO Guillermo Rauch and others praised the coding quality and speed in the first hours. Vercel AI Gateway already added GLM 5.2 Fast via Wafer (170+ tokens/s).

How to try

Weights: HuggingFace under MIT. API: z.ai with the new GLM Coding Plan subscription. Also integrated into Vercel AI Gateway and Replicate. For local runs, suitable for ollama / vLLM.