GLM-5.2 is Zhipu AI's June 13, 2026 flagship: a ~753B-parameter mixture-of-experts model with ~40B active parameters, a 1-million-token context window, up to 128K output tokens, and MIT-licensed weights on Hugging Face. It is built deliberately as a coding and agent model — configurable thinking effort, native tool calling, MCP integration, structured JSON output — rather than a general-purpose chat model.
The engineering hook is IndexShare, Zhipu's sparse-attention refinement that reuses indexers across sparse-attention layers to cut per-token compute at long context. That is what makes the 1M-token window usable in practice for long-horizon coding agents, not just impressive on a spec sheet.
Launch coverage framed GLM-5.2 as matching GPT-5.5-tier coding at roughly one sixth of the cost — continuing the GLM line's pattern (GLM-4.5, 4.6, 5, 5.1) of aggressively pricing against Claude and GPT for the coding-agent market.
GLM-5.2 specifications
| Vendor | Zhipu AI (Z.ai) (China) |
|---|---|
| Architecture | Mixture-of-Experts with IndexShare sparse attention (indexers reused across sparse-attention layers) |
| Total parameters | 753B |
| Active parameters | ~40B |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Modalities | text |
| License | MIT — weights free to download, self-host, fine-tune, and use commercially |
| Release date | 2026-06-13 |
| Weights | Hugging Face · GitHub |
Benchmarks and reported results
| Benchmark | Result | Note |
|---|---|---|
| Positioning | coding-first | launch coverage reports GPT-5.5-tier coding at roughly one sixth the API cost |
Running GLM-5.2 locally
753B total / ~40B active: 4-bit builds land roughly in the 400–450GB range — an 8-GPU A100/H100 node or equivalent. Comfortably hostable by inference providers; beyond desktop rigs.
What GLM-5.2 is best for
- Long-horizon coding agents (1M-token context built for repository-scale work)
- Teams that want an MIT-licensed Claude-for-coding alternative
- Tool-driven agentic workloads (native tool calling, MCP, JSON output)
Frequently asked questions
Is GLM-5.2 open source?
GLM-5.2 is open-weight under the MIT license — free to download from Hugging Face, self-host, fine-tune, and use commercially. As with most frontier releases, the training data itself is not published.
What is GLM-5.2 best at?
Coding and agentic workloads. Zhipu positions it as a coding-first model: 1M-token context for repository-scale tasks, configurable reasoning effort, native tool calling and MCP support, with launch coverage reporting GPT-5.5-tier coding performance at a fraction of the cost.
How does GLM-5.2 relate to GLM-5?
GLM-5.2 builds on the MoE foundation introduced with GLM-5 and GLM-5.1 earlier in 2026, extending the context window to a usable 1M tokens via IndexShare sparse attention while preserving coding performance. It is the current flagship of the GLM-5 family.