Z.ai

What is Z.ai?

Z.ai is a Chinese AI provider that develops the GLM (General Language Model) family and offers its models through a hosted API. Founded in 2019 as a company spun out of Tsinghua University, it provides language, coding, and vision-capable models, with many model weights also released publicly.

On Eden AI, the Z.ai API is available through 11 LLM models, ranging from the lower-cost GLM-4.5-Air to the 1-million-token-context GLM-5.2.

Z.ai and Zhipu AI: the same company

Z.ai is the provider formerly, and still widely, known as Zhipu AI. If you see references to the Zhipu AI API, GLM models, or Z.ai models, they refer to the same company and model family.

The Z.ai name is now used for the provider and its hosted API, while Zhipu AI remains common in older documentation, articles, and developer searches.

Z.ai at a glance

Is Z.ai open source?

Z.ai publishes the weights for many GLM models, with most released under the MIT license. That gives teams the option to self-host supported models rather than depend exclusively on the provider's hosted infrastructure.

The Eden AI integration works differently: when you call a model such as zai/glm-5.2, Eden AI routes the request to the hosted Z.ai API rather than running the open weights itself.

Practically, this gives you two paths. You can use the hosted GLM API for managed inference, or use the published weights for self-hosting where available. With Eden AI, provider lock-in is also reduced at the API layer because the same integration can route requests to other supported AI providers without requiring a separate provider-specific integration.

Z.ai main AI capabilities

• Text Generation APIs: to generate, rewrite or structure text inside applications, evaluated on realistic generative AI inputs.
• Multimodal Chat: to build assistants that can reason across text and other input types.
• Intelligent Chatbot: to power conversational features that stay coherent across long exchanges.
• Code Generation: to write, explain, refactor or review code inside developer workflows.
• Long-Context Reasoning: to process large documents, repositories or conversation histories in a single request.
• Tool Calling and Agents: to let the model trigger functions and chain steps inside an automated workflow.

Z.ai models available on Eden AI

GLM-5.2: the 1M-token flagship

GLM-5.2 is the Z.ai model to use when you need the longest context window available on Eden AI. The zai/glm-5.2 model supports 1.0M tokens and costs $1.40 per 1M input tokens and $4.40 per 1M output tokens. Its clearest fit is long-context reasoning and agent workflows where large documents, codebases, or multi-step task history need to remain in context. It is also the most expensive GLM option in the current Eden AI catalog by output price.

GLM-5.1, GLM-5 and GLM-5-turbo

GLM-5.1, GLM-5 and GLM-5-turbo all support 203K tokens, but they sit at different price points. zai/glm-5.1 costs $1.40 input / $4.40 output per 1M tokens and is the previous flagship. zai/glm-5 costs $1.00 / $3.20 and is the balanced pick for general-purpose LLM traffic. zai/glm-5-turbo costs $1.20 / $4.00 and is the better fit when you specifically want the speed-oriented GLM-5 variant.

GLM-4.7 and GLM-4.6

GLM-4.7 and GLM-4.6 both support 203K tokens and cost $0.60 per 1M input tokens and $2.20 per 1M output tokens. zai/glm-4.7 is positioned as the price-performance option, making it a practical choice for general production workloads where GLM-5 pricing is unnecessary. zai/glm-4.6 is widely used for coding, so it is the more natural pick for code generation, code explanation, and developer-assistant workflows without moving to the higher-priced GLM-5 family.

GLM-4.5 and GLM-4.5-air

GLM-4.5 supports 131K tokens and costs $0.60 input / $2.20 output per 1M tokens. The zai/glm-4.5 model is the earlier-generation option when you need standard text workloads and do not require a 203K or 1M context window. GLM-4.5-Air, available as zai/glm-4.5-air, keeps the same 131K context but drops pricing to $0.20 input / $1.10 output, making it the right pick for low-cost classification, extraction, and summarization traffic.

GLM vision models: GLM-5v-turbo, GLM-4.6v and GLM-4.5v

Z.ai offers three vision-capable models on Eden AI. zai/glm-5v-turbo supports 203K tokens, image and video input, and costs $1.20 input / $4.00 output per 1M tokens. GLM-4.6v costs $0.30 / $0.90 with a 131K context, while GLM-4.5v costs $0.60 / $1.80 with only 66K. Vision therefore comes with a context trade-off: these variants top out below GLM-5.2's 1M context, and GLM-4.5v has the shortest context window in the catalog.

Which GLM model should you choose?

• If you need the cheapest possible text model → use GLM-4.5-Air, because zai/glm-4.5-air costs just $0.20 input and $1.10 output per 1M tokens.
• If you need a coding-focused option → use GLM-4.6, because it is widely used for coding while staying at $0.60 input and $2.20 output per 1M tokens.
• If you need the longest context → use GLM-5.2, because it is the only model in the catalog with a 1.0M-token context window.
• If you need image input → use GLM-4.6v for lower-cost vision or GLM-5v-turbo for the largest vision context, because they support 131K and 203K tokens respectively.
• If you need the best balance between price and capability → use GLM-4.7, because it combines a 203K context window with $0.60 input and $2.20 output pricing.

Z.ai API pricing

Z.ai API pricing is token-based, with input and output charged separately per million tokens. Across the models available on Eden AI, input pricing ranges from $0.20 to $1.40 per 1M tokens, a 7× spread between the cheapest and most expensive options.

GLM pricing per million tokens

Output pricing matters especially for generation-heavy workloads. A model with cheap input can still become expensive if your application consistently produces long responses.

How much does GLM-5.2 cost?

GLM-5.2 costs $1.40 per 1M input tokens and $4.40 per 1M output tokens. For a workload using 1 million input tokens and 200,000 output tokens with zai/glm-5.2, the calculation is:

• Input: 1M × $1.40 = $1.40
• Output: 0.2M × $4.40 = $0.88
• Total: $2.28

You are paying more than for the lower GLM tiers, but GLM-5.2 is also the only model in this catalog with a 1.0M-token context window. That makes the higher unit price easier to justify for long-context reasoning and agent workflows.

Cheapest GLM model: GLM-4.5-air

GLM-4.5-Air is the cheapest text model, at $0.20 per 1M input tokens and $1.10 per 1M output tokens.

For 1 million input tokens plus 200,000 output tokens with zai/glm-4.5-air:

• Input: 1M × $0.20 = $0.20
• Output: 0.2M × $1.10 = $0.22
• Total: $0.42

That is a practical fit for high-volume classification, extraction, and summarization where a 131K context window is sufficient.

Is Z.ai cheaper than OpenAI and Anthropic?

Based on the listed API prices, Z.ai is significantly cheaper than both OpenAI and Anthropic at comparable flagship tiers. GLM-5.2 costs $1.40 input / $4.40 output per 1M tokens, versus $5 / $30 for GPT-5.5 and $5 / $25 for Claude Opus 5. That makes GLM-5.2 about 3.6× cheaper on input and 6.8× cheaper on output than GPT-5.5, and 3.6× cheaper on input and 5.7× cheaper on output than Claude Opus 5.

The gap also exists at the budget tier. GLM-4.5-Air costs $0.20 / $1.10, compared with $1 / $5 for Claude Haiku 4.5, making Z.ai roughly 5× cheaper on input and 4.5× cheaper on output.

Price per token is not the same as cost per completed task, however. For a fair comparison, match GLM models against OpenAI and Anthropic models intended for a similar workload, then compare both input and output rates. Also account for context limits, output quality, retries, fallback calls, and manual correction: a model that costs half as much per token can still cost more overall if it needs repeated calls to produce an acceptable result.

How to reduce Z.ai API costs with model routing

A simple routing policy is to send classification and routine summarization to GLM-4.5-Air, where input costs $0.20 per million tokens. Route long-context reasoning and agent work to GLM-5.2, where the 1M context window is the main reason to accept the higher $1.40/$4.40 pricing.

For workloads between those extremes, GLM-4.7 offers a 203K context window at $0.60 input and $2.20 output per million tokens.

Do not optimize only for the lowest listed token price. If a cheaper model produces more failed outputs, retries, or correction calls, its real cost per completed task can exceed that of a more expensive model that succeeds on the first request.

What can you build with the Z.ai API?

Text generation and content automation

The Z.ai API can generate and transform text for workflows such as product descriptions, report drafts, email generation, and structured summaries. A developer building a CMS feature that turns structured product data into publishable copy could start with GLM-4.7, which provides a 203K-token context window at $0.60 per 1M input tokens and $2.20 per 1M output tokens. 

Before shipping, test output consistency on your own templates and edge cases, especially where formatting or factual fidelity matters.

Chatbots and conversational AI

GLM models can power conversational interfaces where your application sends conversation history and generates the next response. For example, you could build an in-product support assistant that answers questions using account context and previous messages. 

GLM-5 is a sensible starting point because it offers a 203K context window at $1.00 per 1M input tokens and $3.20 per 1M output tokens, positioning it between the cheaper GLM-4 models and higher-priced GLM-5 variants. Before production, test how response quality changes as conversation history becomes longer and more complex.

Code generation and coding agents

You can use GLM models for code generation, explanation, modification, and coding-agent workflows that repeatedly process instructions and source code. A recognizable implementation would be an IDE assistant that receives a file or code section, explains an issue, and proposes a patch. 

Start with GLM-4.6, which is widely used for coding and combines a 203K context window with pricing of $0.60 per 1M input tokens and $2.20 per 1M output tokens. Before shipping, test it against your actual languages, repositories, and coding tasks rather than assuming performance transfers equally across environments.

Multimodal and image understanding

Z.ai's vision models let applications combine visual input with language-model responses. For example, you could build a workflow where a user submits an image and asks the application to describe or analyze what it contains. 

Start with GLM-4.6V if cost is important: zai/glm-4.6v supports a 131K context window and costs $0.30 per 1M input tokens and $0.90 per 1M output tokens. If you also require video input, GLM-5V-Turbo supports image and video. Before shipping, test accuracy on the exact image types your product receives.

Long-context document processing

Long-context processing lets you send substantially more source material in one request before you need to split or reduce it. GLM-5.2 is the model to start with because zai/glm-5.2 supports a 1.0M-token context window, versus 203K or less for every other GLM model listed on Eden AI. 

In practical API terms, that window can hold up to 1 million tokens of combined context, such as large document collections, extensive code, or long task histories. Before shipping, test whether your real inputs fit once instructions and expected output also consume context. 

How to use the Z.ai API

Get your API key

To call Z.ai models through Eden AI, you first need an Eden AI account and API key.

1. Create or sign in to your Eden AI account.
2. Open the API key section in the Eden AI dashboard. 
3. Create a new API key.
4. Copy the key and store it in a server-side environment variable or secret manager.
5. Test the key with a request to a GLM model such as zai/glm-4.6.

You do not need a separate Z.ai or Zhipu AI integration when you access the provider through Eden AI.

Your first Z.ai API request

A basic request needs your Eden AI API key plus the provider and model you want to call. For a first test, GLM-4.6 is a practical starting point because it has a 203K context window and costs $0.60 per 1M input tokens and $2.20 per 1M output tokens.

import requests

response = requests.post(
    "https://api.edenai.run/v3/chat/completions",
    headers={"Authorization": "Bearer <your-api-key>"},
    json={
        "model": "zai/glm-4.6",
        "messages": [
            {"role": "user", "content": "Hello, world!"}
        ]
    }
)

print(response.json())

‍

Switch between GLM models with one parameter

Once the request is working, switching GLM models only requires changing the model string. For example, you can replace zai/glm-4.5-air with zai/glm-5.2 while keeping the rest of the integration unchanged.

That makes model evaluation straightforward. Run the same classification, summarization, coding, or long-context test set against both models and compare output quality, latency, and token cost on your own traffic.

This is particularly useful when deciding whether GLM-5.2's 1M-token context and higher $1.40/$4.40 pricing justify moving traffic away from a lower-cost model.

Authentication and key security

Treat your Eden AI API key as a backend credential. Store it in an environment variable, secret manager, or equivalent server-side configuration rather than hard-coding it into your application.

Never expose the key in browser JavaScript, mobile client code, logs, screenshots, or a public Git repository. If a key is accidentally exposed, rotate or revoke it immediately and replace it in every service that uses it.

Handling responses, errors and rate limits

Do not pass a model response directly into downstream systems without validating it first. Check that the expected response fields are present, confirm the returned content is non-empty, inspect token-usage metadata where available, and check the finish reason before treating a generation as complete.

Your application should also handle authentication failures, invalid model names, malformed requests, provider errors, timeouts, and rate-limit responses explicitly.

For production workloads, log the model used, request outcome, token consumption, and retry count. That gives you enough data to distinguish a genuinely cheaper model from one that reduces unit price but increases failures or repeat calls.

Best Z.ai alternatives on Eden AI

Z.ai vs OpenAI

This is where Z.ai's economics land hardest. GLM-5.2 and GPT-5.5 both offer a 1M-token context window, but GLM-5.2 costs $1.40 input / $4.40 output per 1M tokens against GPT-5.5's $5.00 / $30.00 - roughly 6x cheaper on output, the side that dominates cost in generation-heavy and agentic workloads. 

Independent benchmark coverage has GLM-5.2 matching or beating GPT-5.5 on several long-horizon coding tasks at a fraction of the price. OpenAI keeps the advantage in ecosystem depth, audio and realtime modalities, and US/EU data residency.

Verdict: choose Z.ai for high-volume coding and long-context generation where you can validate output; choose OpenAI when you need modalities beyond text and vision, or hosting outside Asia-Pacific.

Z.ai vs Anthropic

Compare tier for tier rather than flagship to flagship. Claude Opus 5 runs $5.00 / $25.00 per 1M at 1M context; GLM-5.2 delivers the same context window at $1.40 / $4.40. 

The closer fight is at the cheap end: Claude Haiku 4.5 is $1.00 / $5.00 with a 200K window, while GLM-4.7 gives you 203K at $0.60 / $2.20 - under half the output cost at effectively the same context. Anthropic's counter-arguments are consistency across long agent runs, enterprise governance, and availability on Bedrock, Vertex and Foundry with US and EU regions.

Verdict: choose Z.ai when token cost governs how much work you can afford to run; choose Anthropic for agent stability, compliance posture, and hosting flexibility.

Z.ai vs DeepSeek 

Z.ai loses this comparison on price. DeepSeek V4 Pro costs $0.66 per 1M input tokens and $1.98 per 1M output tokens with a 1M-token context window, which is less than half the rate of GLM-5.2 at the same context length. DeepSeek V4 Flash also undercuts GLM-4.5-Air on output, at $0.66 versus $1.10 per 1M tokens.

DeepSeek also offers a 1M-token context window across both of its models, while Z.ai offers 1M context on only one of its eleven models.

Z.ai's advantage is model breadth. You get three vision models - GLM-5V-Turbo, GLM-4.6V, and GLM-4.5V - plus eleven pricing tiers, which gives you more room to route mixed workloads by cost and capability. 

Verdict: DeepSeek wins on price; choose Z.ai when you need vision and text within one model family or finer-grained model routing across different workloads.

Z.ai vs Mistral AI

Mistral is the stronger default for any European team, and not only on hosting. Mistral Small 4 is $0.15 / $0.60 per 1M with a 262K context - cheaper than GLM-4.5-Air on both directions with twice the window - and Mistral Large 3 is $0.50 / $1.50 at 262K, fully open weights under Apache 2.0, served from the EU. 

Z.ai's edge is the top of the range: GLM-5.2's 1M context is roughly 4x Mistral's ceiling, which matters for whole-repository or full-corpus requests that will not fit in 262K.

Verdict: choose Mistral when EU data residency applies at all, or for cost-sensitive work under 262K tokens; choose Z.ai when a single request genuinely needs 1M tokens of context.

Why access Z.ai through Eden AI

One API for Z.ai and 60+ providers

Z.ai is a strong option for low-cost GLM models and long-context workloads, but on Eden AI it is hosted in Asia-Pacific only. If you later need an EU-hosted model for data residency, lower European latency, or a customer-specific deployment constraint, a direct Z.ai integration would normally force you to add a second provider API. 

Through Eden AI, the application keeps the same API surface while you change the model or provider route, so adding an alternative does not require rebuilding authentication, request handling, and response parsing.

Fallback and routing for reliability

A production system should not assume one model is the right choice for every request. You might route routine summarization to zai/glm-4.5-air, coding traffic to zai/glm-4.6, and long-context agent requests to zai/glm-5.2, then send specific workloads to another provider when Z.ai's AP-only hosting conflicts with latency or residency requirements. 

The same routing layer can also support fallback when a selected provider is unavailable or unsuitable, without maintaining separate integration logic for every model family.

Monitoring, usage and billing in one place

Model routing becomes harder to optimize when cost and usage data are split across separate provider dashboards. Z.ai alone spans 11 models, from $0.20 per 1M input tokens for GLM-4.5-Air to $1.40 for GLM-5.2, so even one provider creates multiple cost profiles to track. 

If you benchmark those models against OpenAI, Anthropic, DeepSeek, or Mistral, centralized usage and billing make it easier to compare token consumption, model selection, and spend without reconciling separate accounts and invoices.

Use the Z.ai API through Eden AI when you want to test GLM models alongside other providers without building a separate integration for each one.

Z.ai API FAQ