All AI models
Open sourceVerified August 11, 2026Previous gen

Z.ai

GLM-5.2

GLM-5.2 is worth evaluating for long-horizon coding and agent work where one million tokens of stable context and an MIT license matter. It is a frontier-scale open model, so the real decision is usually between a managed plan and purpose-built serving infrastructure, not whether to run it on a personal computer.

What changed: Newer from Z.ai: GLM-5.3, released August 18, 2026, lists the same $1.40 / $4.40 price with a 1M-token context window and 128K max output. On Mistral's platform, the hosted zai-glm-5-2 model was deprecated on September 29, 2026; as of October 23, 2026, Mistral lists its retirement for October 31, 2026.
License
MIT
Context
1M tokens
Modalities
Text
Parameters
744B total, 40B active

What the license means

Z.ai describes GLM-5.2 as a pure open release under an MIT license and publishes weights through Hugging Face and ModelScope. Read the selected weight repository before deploying an altered or redistributed version.

Open-weight, open source, and no-cost API access mean different things. Always review the linked terms and your own data, hosting, and compliance requirements before shipping.

API and operating cost

Input / 1M
$1.40
Output / 1M
$4.40

Z.ai pricing page, per 1M tokens, as listed on October 23, 2026. GLM-5.3 and GLM-5.1 list the same rate; GLM-5 costs $1 / $3.20.

Check current provider price

Where it fits

  • Long-horizon coding agents
  • Large-scale repository reasoning
  • Self-hosted enterprise experimentation

Where it does not fit

  • Laptop deployment
  • New projects that could start on GLM-5.3 at the same API price
  • Vision workloads

Local deployment reality

Z.ai documents Transformers, vLLM, SGLang, xLLM, and KTransformers. The 744B total parameter footprint means local use is a data-center or serious multi-GPU decision, even though the model is openly available.

TransformersvLLMSGLangxLLMKTransformers

Practical setup notes

  • Choose thinking effort deliberately for latency-sensitive tasks.
  • Use a supported serving runtime and measure cache capacity before designing around the full context window.

Capability notes

Reasoning
Flexible thinking effort levels
Tools and output
Function calling, structured output, context caching, coding plan integration

Limitations to plan around

  • The context window does not make one-million-token prompts inexpensive to serve.
  • Z.ai has since released GLM-5.3, so check where new features land before standardizing on GLM-5.2.

Continue comparing

Related model decisions

See all reviewed models
Open-weight

Kimi K3

Long-horizon coding agents

Open-weight

DeepSeek V4

Cost-sensitive API workflows

Open-weight

MiniMax M3

Agentic coding

Editorial note: This is a dated decision page, not a benchmark leaderboard or a substitute for legal, security, or infrastructure review. Model and pricing facts can change after August 11, 2026; use the linked primary sources before committing budget or customer data.API price rechecked on October 23, 2026. License, parameter, and deployment notes date from the August 11, 2026 audit.