Z.ai releases GLM-5.2 open-weights model that beats GPT-5.5 on coding benchmarks at one-sixth the cost

Zhipu AI's Z.ai released GLM-5.2 on June 13, 2026, a 753-billion-parameter mixture-of-experts model that outperforms GPT-5.5 on multiple long-horizon coding benchmarks while costing roughly one-sixth as much to run via API. The model weights are published under the MIT license — no usage restrictions, no regional locks.

What's new

GLM-5.2 is Z.ai's new flagship open-weights model, succeeding GLM-5.1 with a dramatically expanded context window and stronger performance on agentic coding tasks.

Key specifications:

• Architecture: Mixture-of-experts, 753B total / ~40B active parameters
• Context window: 1 million tokens (up from 200K in GLM-5.1)
• Max output: 128K tokens
• License: MIT (weights downloadable from Hugging Face)
• Capabilities: Thinking mode, function calling, context caching, structured output, MCP integration

Benchmark results:

• FrontierSWE: Edges out GPT-5.5 by approximately 1%; trails Claude Opus 4.8 by 1%
• SWE-bench Pro: 62.1 (up from GLM-5.1's 58.4)
• Terminal-Bench 2.1: 81.0 (vs. Claude Opus 4.8 at 85.0)
• MCP-Atlas (tool use): Outperforms GPT-5.5
• Humanity's Last Exam (with tools): Outperforms GPT-5.5

Pricing via OpenRouter: approximately $1.40 per million input tokens and $4.40 per million output tokens — compared to $5/$30 for GPT-5.5 and $5/$25 for Claude Opus 4.8. Z.ai also launched a GLM Coding Plan subscription starting at $12.60 per month for agent workloads.

GLM-5.2 is available through the Z.ai API, Hugging Face, and more than 20 third-party coding environments.

Context

Zhipu AI has released open-weight models through its Z.ai brand since 2024. GLM-5 arrived in early 2026 focused on coding and agentic tasks; GLM-5.1 extended long-horizon capabilities. GLM-5.2 continues that trajectory, specifically targeting the gap between closed-source frontier performance and open-weight accessibility.

The MIT license stands out. Several high-performing open models from Chinese labs have carried commercial restrictions or geographic limitations. GLM-5.2 carries neither. Enterprises can download the weights, fine-tune, and run inference on their own infrastructure without restriction.

The release coincides with a period of disruption in frontier model access: the US government issued an export control directive on June 12 suspending access to Fable 5 and Mythos 5 for foreign users, adding urgency to the open-weights alternative.

Why it matters

GLM-5.2 is the most capable open-weights coding model currently available by most independent measures. That matters for enterprises and developers who cannot route code through closed-source API endpoints — regulated industries, security-conscious organizations, and teams running inference on their own hardware.

At one-sixth the API cost of GPT-5.5, GLM-5.2 makes frontier-grade coding assistance far more accessible to smaller teams. The 1M-token context window means it can ingest a full production codebase in a single pass, a capability few open models have matched at this scale.

The release also marks continued momentum from Chinese AI labs in the open-weights space, at a moment when US export controls are reshaping international access to American frontier models.