GLM-5.3 Explained: Price, Specs and Open Weights

In short: GLM-5.3 is a large language model from Z.ai (Zhipu AI), released on 18 August 2026. It uses the same base model as GLM-5.2, and Z.ai attributes all of its improvements to post-training. It costs $1.40 per million input tokens and $4.40 per million output tokens through Z.ai's API, has a 1M-token context window and 128K tokens of output, reads text only, always reasons before answering, and its weights are published on Hugging Face.

What is GLM-5.3?

GLM-5.3 is Z.ai's flagship model, built for coding and agent tasks. Z.ai's model documentation describes it as using the same base model as GLM-5.2, "with all improvements driven by post-training" - extra training after the main pre-training stage, rather than a new model trained from scratch. Z.ai is the international brand of Zhipu AI, a Chinese AI company. The API model ID is glm-5.3.

For what pre-training and post-training are, see how LLMs work.

How much does GLM-5.3 cost?

Input costs $1.40 per million tokens and output $4.40, the same as GLM-5.2. From Z.ai's pricing page:

Reasoning tokens are billed as output, so higher effort settings cost more per request. GLM-5.3 is also included in Z.ai's GLM Coding Plan subscription.

What are GLM-5.3's specs?

The parameter count comes from Z.ai's GitHub repository, which lists GLM-5.3 and GLM-5.2 as 744B-A40B: 744 billion parameters in total, of which 40 billion are used for each token - a mixture-of-experts design.

Can you turn off reasoning on GLM-5.3?

No. Z.ai's documentation says GLM-5.3 always operates with reasoning enabled and that disabling reasoning is no longer supported. It offers three effort levels - low, high and max, with max as the default. Applications that used thinking.type: "disabled" on earlier GLM models have to switch to enabled with reasoning_effort: "low" before changing the model ID, or the request fails.

What did Z.ai report about GLM-5.3's performance?

Z.ai published benchmark results against GLM-5.2; these are the vendor's own figures, not independent tests. From the GLM-5.3 model card:

Z.ai also reports a 50% improvement over GLM-5.2 on its in-house Z.ai Code Bench. Independent rankings are on our benchmarks page.

Is there a smaller GLM-5.3?

Yes - Z.ai also released GLM-5.3-Flash and GLM-5.3-FlashX. Z.ai's repository lists GLM-5.3-Flash at 320B total and 18B active parameters, built on a newly trained base model with a hybrid of sparse and linear attention. On Z.ai's pricing page, GLM-5.3-Flash costs $0.15 per million input tokens and $0.50 per million output, and GLM-5.3-FlashX $0.37 and $1.25.

How does GLM-5.3 rank?

On LMArena's text leaderboard GLM-5.3 has 1481 Elo (rank 28, max effort). GLM-5.2 has 1475 Elo (rank 38, max effort). The full ranked table is on our benchmarks page.

Words to know

• Post-training - training done after the main pre-training stage, such as fine-tuning and reinforcement learning.
• Base model - the model produced by pre-training, before post-training.
• Mixture of experts (MoE) - a design that uses only part of the model for each token.
• Active parameters - the parameters actually used for each token.
• Reasoning effort - how much the model thinks before answering.
• Open weights - model files published for download.

Related guides

• How LLMs work
• How to run AI models locally
• How to compare AI models
• What is a model card?