> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tess.im/llms.txt
> Use this file to discover all available pages before exploring further.

# GLM 5.2

The GLM-5.2 is the high-performance *open-source* language model developed by Zhipu AI, designed specifically for *long-horizon tasks*.

With a massive context window of 1 million tokens with no loss of accuracy (*lossless*), it is the ideal choice for reading complete codebases, complex interconnected refactoring, and advanced development of agent-structured engineering *workflows* in Tess.

| **Model ID**<br /><br />glm-5.2                                           | **Context Window**<br /><br />1M | **Max Context**<br /><br />128K | **Provider**<br /><br />Zhipu AI         |
| :------------------------------------------------------------------------ | :------------------------------- | :------------------------------ | :--------------------------------------- |
| **Capabilities**<br /><br /><Icon icon="infinity" /><Icon icon="brain" /> | **Speed**<br /><br />Medium      | **Cost**<br /><br />Medium      | **Intelligence**<br /><br />Text-to-Text |

## Capabilities

* Reasoning: Has *Thinking* mode with adjustable capacity and controllable effort for difficult tasks.
* Tools (Function Calling / MCP): Highly reliable and focused on staying within scope.
* Structured Output (JSON): Ideal for seamlessly integrating with the operational flows of external tools.

<Info>
  See more in the official documentation: [glm 5.2 documentation](https://docs.z.ai/guides/llm/glm-5.2).
</Info>

## Details (context, cost, retention)

<Card title="Extreme Lossless Context">
  Support for 1 million tokens makes it possible to send an entire code repository, extensive manuals, and heavy histories in a single agent *prompt*. It keeps continuous information without getting lost in "hallucinations" as the chat progresses. It supports *Context Caching*, which makes long conversations cheaper on the platform.

  <Note>
    To use the maximum context, you need to activate Max Mode in the chat, but this may involve higher costs.
  </Note>
</Card>

<Card title="Cost Efficiency">
  It can outperform several competitors (such as DeepSeek v4 and Gemini 3.1 Pro in sustained coding) while costing around 1/6 of the price of equivalent proprietary alternatives from other providers.
</Card>

## Pricing and consumption

Credit consumption in Tess for this model occurs according to the tokens processed:

* Input Tokens (Environmental reading / *Prompt*): 0.672 credits / 1K tokens
* Output Tokens (Response generation): 2.112 credits / 1K tokens

<Note>
  *Tasks with 1M tokens can generate a high peak in reading consumption due to the absolute volume of data entered in the input. Using the Context Caching feature can help automatically reduce this cost.*
</Note>

<Tip>
  **Best practices**

  * **Set safety limits in coding:** Because it strictly follows production architectural standards, give clear restrictive instructions in the *prompt*, such as: *"Adopt the lint standards, use commit convention X, and test rule Y in isolation"*. GLM-5.2 retains this command much better than conventional models.
  * **Increase reasoning in interactions with *Bugs*:** For problems such as server log analysis, instruct it in your *prompt* and in the Agents to use step-by-step reasoning before printing the final solution.
  * **Redirect multimodal cases:** Since it does not have image capabilities (*Vision*), if your automation needs to read screens and run tests using screenshots of the visual interface, route this step first to the GLM-5V-Turbo or GLM-OCR models.
</Tip>

The GLM-5.2 model by Zhipu AI breaks the barrier between open source and top-tier enterprise execution. With its ability to absorb large amounts of information combined with logical reasoning focused on staying on track for long tasks, it is the ideal tool inside Tess for engineers, advanced researchers, and automation creators who cannot risk technical failures across large volumes of data transition.