Zhipu Z.ai GLM Models: Every Model, Spec, Price and Release Date

GLM is the family of large language models from Zhipu, offered internationally under the Z.ai brand through the Z.ai API Platform and as open weights on Hugging Face; its current flagship, GLM-5.3, was released on 18 August 2026 with a 1M token context window. This page lists every current GLM model, its API identifier, context window, licence, published price and status, plus the release timeline from GLM-4.6 to GLM-5.3, the lifecycle position of older models and an index of model cards and technical reports. Every fact comes from Z.ai's own documentation, release notes, pricing page, legal pages or Hugging Face model cards. Where Z.ai publishes no value, the page says "Not stated by Z.ai" rather than estimating. Facts are as of 15 November 2026.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.

Key takeaways

• GLM-5.3 is the flagship. Released 18 August 2026, API name glm-5.3, 1M context, 128K maximum output, text-only input and 753B total parameters on its Hugging Face card; Z.ai describes it as using the same base model as GLM-5.2 with gains from post-training.
• GLM-5.3-Flash is the first natively multimodal GLM-5 model. Released 26 August 2026, it accepts video, image, text and file input, uses 320B total and 18B active parameters and is published under the MIT licence.
• Prices are published in US dollars. GLM-5.3 costs $1.4 input, $0.26 cached input and $4.4 output per million tokens; GLM-5.3-Flash costs $0.15, $0.03 and $0.50, and cached input storage is listed as limited-time free.
• Reasoning is always on for GLM-5.3. Requests that set thinking.type to disabled fail; Z.ai's migration note says to use enabled with reasoning_effort low instead.
• Older models stay on the price list. GLM-5.2, GLM-5.1, GLM-5 and the GLM-4.x models remain priced, and three Flash models (GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash) are listed as free.

What is Zhipu Z.ai?

Z.ai publishes the GLM models through the Z.ai API Platform (docs.z.ai), the GLM Coding Plan subscription for coding tools and the zai-org organisation on Hugging Face. Z.ai's privacy policy names JINGSHENG HENGXING TECHNOLOGY PTE. LTD in Singapore as the data controller for individual users and states that personal data and API customer data are generally processed in Singapore.

Model names follow a generation and point-release pattern: GLM-4.5, 4.6, 4.7, then GLM-5, 5.1, 5.2 and 5.3. Suffixes mark a variant: Flash for a smaller, lower-cost model, FlashX for a faster Flash tier, Air and AirX for lighter GLM-4.5 models, V for vision models and Legal for a legal-domain variant. API identifiers are lower-case versions of the names (glm-5.3, glm-5.3-flash, glm-5.3-flashx, glm-5.3-flashx-legal). See LLM Naming and Versioning Explained for how these conventions work across providers.

Current GLM models at a glance

As of 15 November 2026. Sources: Z.ai pricing page, model guides and Hugging Face model cards. Models are listed in the order of Z.ai's pricing page.

The pricing page also lists GLM-4.7-FlashX, GLM-4.5, GLM-4.5-X, GLM-4.5-Air, GLM-4.5-AirX, GLM-4-32B-0414-128K, GLM-4.6V-FlashX and the free Flash models, plus non-token products: GLM-Image, CogView-4, CogVideoX-3, GLM-ASR-2512 and agent services.

Key facts for the main GLM models

As of 15 November 2026. Identical fields for every model; "Not stated by Z.ai" means not stated by Z.ai on the pages checked (2026-11-15). API IDs, context, licence and status are in the table above.

Release dates come from Z.ai's release notes. GLM-5.3-Flash uses a hybrid sparse and linear attention architecture, which Z.ai says reduces serving costs, and GLM-5 roughly doubled the parameter count of the previous generation (355B total, 32B active). These models use a Mixture-of-Experts design, which is why the cards give total and active counts. See Mixture of Experts Explained.

API capabilities listed in the Z.ai model guides

GLM-5.3-FlashX is listed at inference speeds of 200 tokens per second and was not yet available on the GLM Coding Plan when checked. GLM-5.2 was available early to GLM Coding Plan users before its official release. See Reasoning Models Explained.

GLM model families and naming

Z.ai's model lines, as of 15 November 2026, from the pricing page and release notes.

• GLM-5 line: GLM-5 (February 2026), GLM-5.1 (April), GLM-5.2 (June), GLM-5.3 (August). GLM-5.3 shares its base model with GLM-5.2; the GLM-5 technical report, arXiv 2602.15763, is cited by the GLM-5, 5.2 and 5.3 cards.
• GLM-5.3 Flash tier: GLM-5.3-Flash, GLM-5.3-FlashX (faster) and GLM-5.3-FlashX-Legal (legal document understanding, contract review and drafting).
• GLM-4 line: GLM-4.5 with X, Air, AirX and Flash variants, GLM-4.6, GLM-4.7 with FlashX and Flash variants, and GLM-4-32B-0414-128K.
• Vision and document: GLM-4.5V, GLM-4.6V with FlashX and Flash variants, and GLM-OCR.
• Generation models: GLM-Image (released 2026-01-14), CogView-4 for images, CogVideoX-3 for video and GLM-ASR-2512 for speech recognition.

When was each GLM model released?

Dates from Z.ai's release notes, as of 15 November 2026, in date order.

The release notes do not list GLM-5.3-FlashX or GLM-5.3-FlashX-Legal; the GLM-5.3-Flash guide describes both as live without a date.

How much do GLM models cost?

As of 15 November 2026. USD per million tokens. Source: Z.ai pricing page (docs.z.ai/guides/overview/pricing).

Other published prices: GLM-4.6V costs $0.3 input and $0.9 output, GLM-OCR $0.03 input and output, GLM-4.5-X $2.2 and $8.9, and GLM-4-32B-0414-128K $0.1 and $0.1 per million tokens. GLM-Image is $0.015 per image, CogVideoX-3 $0.2 per video, GLM-ASR-2512 $0.03 per million tokens, and the built-in Web Search tool $0.01 per use. The "limited-time free" label on cached input storage has no end date on the page; confirm current prices in the console before committing budgets.

For how input, output and cached pricing work in general, see Model Pricing Explained. To cap spend on keys, see How to set spend limits on LLM API keys.

Where are GLM models available?

• First-party API: the Z.ai API Platform, using the model codes above.
• GLM Coding Plan: a subscription for coding tools; GLM-5.3 is available on it, while GLM-5.3-FlashX was not yet available on it when checked.
• Open weights: GLM-5.3, GLM-5.3-Flash, GLM-5.2 and GLM-5 are on Hugging Face under zai-org.
• Self-hosting engines named on the GLM-5.3 card: SGLang, vLLM, TokenSpeed, Hugging Face Transformers, KTransformers, Unsloth, Ascend NPU options (vLLM-Ascend, xLLM, SGLang) and Docker Model Runner, with quantised community builds listed for llama.cpp, Ollama and LM Studio.
• Third-party clouds and regions: not stated by Z.ai on the pages checked (2026-11-15). Hosts that serve GLM models document them on their own pages, for example Mistral AI Models.

Running open weights yourself moves security responsibility to you. See the Open-Weight Model Security Guide and the Self-Hosted LLM Security Guide.

Restricted-access GLM models

As of 2026-11-15. Restricted-access research programmes for GLM models: not stated by Z.ai on the pages checked (2026-11-15).

What is the context window of GLM models?

GLM-5.3, GLM-5.3-Flash (with FlashX) and GLM-5.2 have a 1M token context window and 128K maximum output in Z.ai's model guides. GLM-5.1 and GLM-5 have 200K context and 128K maximum output. The Hugging Face cards quote 128K, 300K and 400K figures in benchmark footnotes; those are evaluation settings, not model limits. See Context Windows Explained and LLM Context Windows Compared Across Providers.

Is any GLM model open source?

GLM-5.3-Flash, GLM-5.2 and GLM-5 publish weights under the MIT licence, which the GLM-5.2 card calls "an MIT open-source license". GLM-5.3 publishes weights under a licence named "glm-5.3" on its Hugging Face card; read its terms before commercial use. Open weights for GLM-5.1, GLM-5.3-FlashX and FlashX-Legal are not stated by Z.ai on the pages checked. Open weights are not the same as open training data. See Open-Weight vs Closed Models Explained.

Model cards and system cards

Z.ai's model guides are specification and usage pages (context, output, modalities, capabilities, request examples). The GLM-5.3 guide carries a migration note for code that turned thinking off. See Model Cards and System Cards Explained.

Benchmarks

As of 15 November 2026. Provider-reported scores from the Hugging Face model cards. Independent results are kept in a separate column.

Rows marked "model guide" come from Z.ai's GLM-5.3 guide rather than the Hugging Face card. The GLM-5.3-Flash card also reports Real5-OmniDocBench overall 90.76, a document parsing result with no GLM-5.3 equivalent. Z.ai's GLM-5.3 guide also reports an in-house Z.ai Code Bench result of 34.5% at max effort against 23.4% for GLM-5.2. In-house benchmarks cannot be reproduced outside the provider. See How to Read LLM Benchmarks.

Deprecations and retirements

As of 15 November 2026. Z.ai's release notes and pricing page list no deprecation or retirement dates for GLM models on the pages checked (2026-11-15).

Z.ai keeps superseded models on sale alongside new ones: GLM-5.2 and GLM-5.1 carry the same price as GLM-5.3, and GLM-4.5 variants remain listed more than a year after GLM-4.6 was released. Record which model code each workload uses so a later retirement notice can be acted on quickly. One behaviour change is documented: GLM-5.3 no longer accepts thinking.type disabled, so code written for GLM-5.1 that turns thinking off must change when it moves to GLM-5.3. See the AI Model Retirement Guide and the LLM Deprecation and Retirement Tracker.

Data handling and security

• Scope of terms: Z.ai's privacy policy applies to individual users; API customers are covered by a separate Data Processing Addendum in which the customer is the data controller and Z.ai is the processor.
• API content retention: the DPA summary states that the company does not store any of the content the customer or its end users provide or generate, and processes it in real time. Other API data is stored temporarily and deleted after the terms end, unless law requires otherwise.
• Training use: the policy's training purposes list account, communication, log, usage, device and cookie data, and public information; it does not list user content. A training opt-out is not described.
• Individual users: account information and input are retained as long as the account exists; deleted conversations are removed from history and the back end, and account deletion is permanent.
• Location: personal data and API customer data are generally processed in Singapore.
• Key handling: generic controls apply to every provider's API: keep keys out of code, scope and rotate them, and set spend limits.

See the LLMjacking Guide and the Foundation Model Selection Guide. For coding use through the GLM Coding Plan or any other harness, see the AI Coding Agents Security Guide.

How to choose a GLM model: factual criteria

• Input type: GLM-5.3, 5.2, 5.1 and 5 accept text only; GLM-5.3-Flash and FlashX accept video, image, text and file input.
• Licence: MIT for GLM-5.3-Flash, GLM-5.2 and GLM-5; the "glm-5.3" licence for GLM-5.3.
• Price per token: $1.4 and $4.4 for GLM-5.3, $0.37 and $1.25 for FlashX, $0.15 and $0.50 for Flash, and free GLM-4.x Flash models.
• Speed: GLM-5.3-FlashX is listed at 200 tokens per second.
• Context: 1M on GLM-5.3, 5.3-Flash and 5.2; 200K on GLM-5.1 and GLM-5.
• Reasoning control: GLM-5.3 always reasons; Z.ai recommends max effort for complex tasks such as coding, and low is the documented setting for code that previously turned thinking off.
• Older-generation alternatives: GLM-4.7 and GLM-4.6 at $0.6 input and $2.2 output, GLM-4.5-Air at $0.2 and $1.1, and GLM-4.7-FlashX at $0.07 and $0.4 remain on the price list.
• Domain access: GLM-5.3-FlashX-Legal requires an enterprise application.

Practitioner checklist

• Confirm the exact model code (for example glm-5.3 or glm-5.3-flash) in the Z.ai docs before you build against it.
• Check the price per million input and output tokens on the Z.ai pricing page, and note the date you checked.
• Remove thinking.type disabled from requests before moving to GLM-5.3, and set reasoning_effort explicitly.
• Check the context window for the specific model: 1M for GLM-5.3 and 5.2, 200K for GLM-5.1 and GLM-5.
• Read the GLM-5.3 licence file before commercial self-hosting.
• Review the Data Processing Addendum for API use and the processing location (Singapore) against your data residency requirements.
• Watch the pricing page for an end date to limited-time free cached input storage.
• Store Z.ai API keys in a secrets manager and rotate them; never embed them in client code.

Standards and references

• Z.ai: Pricing (accessed 2026-11-15)
• Z.ai: Release notes (accessed 2026-11-15)
• Z.ai: GLM-5.3 model guide (accessed 2026-11-15)
• Z.ai: GLM-5.3-Flash model guide (accessed 2026-11-15)
• Z.ai: GLM-5.3 model card, with the GLM-5.3-Flash, GLM-5.2 and GLM-5 cards (accessed 2026-11-15)
• Z.ai: GLM-5.2 model guide (accessed 2026-11-15)
• Z.ai: GLM-5.1 model guide (accessed 2026-11-15)
• Z.ai: GLM-5 model guide (accessed 2026-11-15)
• Z.ai: Privacy policy (accessed 2026-11-15)

Related NHI Mgmt Group resources: Moonshot AI Kimi Models, Alibaba Qwen Models, LLM Security Guide, Open-Weight Model Security Guide.