Z.ai is the international home of the GLM model family: a free chat assistant, a coding subscription that plugs into Claude Code and Cline, and a pay-per-token API with OpenAI-compatible endpoints. Our review worked through the published pricing, plan credits and usage rules to show what the platform really costs.
📋 Quick Verdict
Z.ai is three products wearing one domain. The consumer chat app is the entry point, the GLM Coding Plan is the value proposition at $18 to $160 per month, and the API is the developer route at $0.15 to $4.40 per million tokens. All three run on the same GLM family, from the Flash models that cost nothing to the flagship GLM-5.3.
Our review found the subscription is the most interesting part. The coding plan targets agent tools rather than the browser, supports Claude Code, Cline, OpenCode and several other harnesses, and grants Vision, Web Search, Web Reader and Zread over MCP on every tier. Credits refresh on rolling five-hour and weekly windows, with Lite starting at 2,000 and 10,000 credits for $18 a month, and off-peak usage is billed at half the credit rate.
The honest caveats are structural. Subscriptions are non-refundable once purchased, calls are restricted to officially supported tools, account sharing is prohibited, and the platform reserves the right to rate-limit or freeze accounts for usage-rule breaches. There is also company-level risk: the firm joined the US Entity List in January 2025, and in February 2026 a compute shortage pushed its share price down 23% and forced it to restrict new signups.
Pros
- Coding plan from $18 per month, working inside Claude Code, Cline, OpenCode, Roo Code, Kilo Code, Goose, Crush, Cursor, Windsurf and Trae
- Frontier token pricing: GLM-5.3 at $1.40 input and $4.40 output per million tokens, with cached input at $0.26
- Flash models are free, and GLM-5.3-Flash costs $0.15 input and $0.50 output per million tokens
- Open weights under the MIT licence since July 2025, so models can be self-hosted if you have the hardware
- Off-peak credit rates are half price, which the docs say can cut costs up to 92% versus pay-as-you-go GLM-5.3
- Every tier includes Vision, Web Search, Web Reader and Zread MCP servers, plus official Python and Java SDKs
- OpenAI-compatible endpoints at api.z.ai/api/paas/v4 make migration a base URL change for most codebases
Cons
- Subscriptions are non-refundable once purchased, even if the quota goes unused
- Plan usage is limited to supported tools; calls from other harnesses can have benefits restricted
- Account sharing is banned, and repeated usage-rule violations risk rate limits or bans
- Credit accounting is more involved than a request count, combining input, cached input, output and MCP multipliers
- The company has been on the US Entity List since January 2025, and a February 2026 compute shortage restricted new signups
- Peak-hour billing doubles the effective credit cost, and peak runs during Asian afternoons and European late mornings
Frontier GLM-5.3 access inside the coding tools you already use, from $18 a month.
🤔 What Is Z.ai?
Z.ai is the international brand of Z.AI Co., Ltd., formerly known as Zhipu AI outside China. The company was founded in Beijing in 2019 by Jie Tang as a Tsinghua University spin-out, and its researchers published the GLM training algorithm at an ACL conference in May 2022. It raised 2.5 billion yuan from investors including Alibaba, Tencent, Meituan and Xiaomi, rebranded internationally as Z.ai in July 2025 with GLM-4.5, and listed on the Hong Kong Stock Exchange on 8 January 2026 as the first major Chinese LLM company to go public. Its market capitalisation was reported at around US$62 billion in August 2026.
As a platform it bundles a consumer chat and agent product, the GLM Coding Plan subscription, and a metered developer API. The model line spans text, vision, image, video and speech, plus agents such as a slide and poster generator and a translation agent. Everything is documented at docs.z.ai with SDKs, API references and a Discord community.
⚙️ Key Features
The GLM model family and open weights
The current flagship is GLM-5.3, with the multimodal GLM-5.3-Flash as the cheap workhorse and older tiers such as GLM-4.7, GLM-4.5-Air and the free Flash models kept available for cost control. Since July 2025 the models have shipped under the MIT licence, so the weights can be downloaded and self-hosted, which is a genuinely different proposition from closed providers and matters most to teams with data-residency constraints or their own GPUs.
GLM Coding Plan for agent tools
This is the subscription most buyers will want. It is built for AI coding agents rather than chat, and the supported list includes Claude Code, Cline, OpenCode, Roo Code, Kilo Code, Goose, Crush, Cursor, Windsurf and Trae. All plans run GLM-5.3 and GLM-5.3-Flash, with requests for GLM-5.2 and GLM-5.1 routed to GLM-5.3 and GLM-4.7 requests routed to GLM-5.3-Flash. Every tier unlocks the Vision, Web Search, Web Reader and Zread MCP servers, which is what lets an assistant read a repository or search the web mid-task.
Credits, rate limits and off-peak pricing
The plan sells credits rather than requests. Usage combines input tokens, cached input tokens, output tokens and MCP calls, each weighted by a model multiplier, then divided by 10,000. GLM-5.3 carries multipliers of 6.9, 1.7 and 24, while GLM-5.3-Flash uses 2.3, 0.56 and 8. Each plan has a five-hour and a weekly credit ceiling, and off-peak usage is charged at half the standard rate. Peak is Monday to Friday, 14:00 to 18:00 Singapore time.
Developer API and SDKs
The API lives at api.z.ai/api/paas/v4 and is OpenAI-compatible, so pointing an existing OpenAI client at that base URL is usually the whole migration. Official SDKs exist for Python (zai-sdk 0.2.3) and Java (ai.z.openapi:zai-sdk 0.3.5), and the OpenAI Python, Node and Java SDKs also work. Documented capabilities include streaming, thinking mode, function calling, structured output, context caching and web search as a tool. Billing is prepaid through the console.
Multimodal and agent add-ons
Beyond text, Z.ai sells media models at per-item prices that are easy to reason about: GLM-Image at $0.015 per image, CogView-4 at $0.01 per image, CogVideoX-3 at $0.20 per video, GLM-OCR and GLM-ASR-2512 at $0.03 per million tokens, and web search at $0.01 per call. Agents include a slide and poster generator, a translation agent and video effect templates, all billed on the same prepaid balance.
💰 Pricing & Plans
Z.ai publishes all model prices in US dollars per million tokens, which makes the API side unusually transparent. The table below is drawn directly from the official pricing page, and the cheapest credible option for most workloads is usually GLM-5.3-Flash rather than the flagship.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GLM-5.3 (flagship) | $1.40 | $0.26 | $4.40 |
| GLM-5.3-Flash (multimodal) | $0.15 | $0.03 | $0.50 |
| GLM-4.7 | $0.60 | $0.11 | $2.20 |
| GLM-4.5-Air | $0.20 | $0.03 | $1.10 |
| GLM-4.7-Flash / GLM-4.5-Flash | Free | Free | Free |
On the subscription side, the GLM Coding Plan has three tiers. Monthly prices are $18 for Lite, $72 for Pro and $160 for Max; quarterly totals are $48.60, $194.40 and $432; annual totals are $172.80, $691.20 and $1,536. Each tier carries a five-hour and weekly credit allowance: Lite 2,000 and 10,000, Pro 12,000 and 60,000, Max 28,000 and 140,000. The five-hour allowance refreshes dynamically after consumption, and the weekly allowance resets every seven days.
| Plan | Monthly | Annual | 5-hour credits | Weekly credits |
|---|---|---|---|---|
| Lite | $18 | $172.80 | 2,000 | 10,000 |
| Pro | $72 | $691.20 | 12,000 | 60,000 |
| Max | $160 | $1,536 | 28,000 | 140,000 |
Two pieces of fine print matter more than the headline numbers. Subscriptions are non-refundable once purchased, and cancellations must be made at least 24 hours before the next billing date to avoid auto-renewal. Plan quota also only applies inside supported tools, and reaching the limit does not silently bill your account balance. If cost predictability is your priority that combination is reassuring; if you expect a refund because you barely used the plan, it is not.
⚔️ Z.ai vs. Competitors
This comparison sticks to structural differences, because closed-model prices change frequently and quoting them as fact would mislead. Verified figures appear only in the Z.ai column; treat the others as descriptions of each provider and check current pricing on their sites.
| Dimension | Z.ai (GLM) | ChatGPT / OpenAI | Claude / Anthropic | DeepSeek |
|---|---|---|---|---|
| Weights | Open, MIT licence since July 2025 | Closed, API and app only | Closed, API and app only | Open weights published |
| Entry cost | Free Flash models and a $18/month coding plan | Free tier plus paid subscriptions | Free tier plus paid subscriptions | Low-cost API, free chat app |
| Agent tool support | Claude Code, Cline, OpenCode, Cursor, Windsurf and more | Own app, API and coding CLI | Own app, API, Claude Code | Own app, API and OpenAI-compatible endpoints |
| Data and compliance | Chinese company, on the US Entity List since January 2025 | US provider, enterprise data controls | US provider, enterprise data controls | Chinese company, self-hosting possible |
| Best fit | Cost-sensitive teams wanting open weights and IDE integration | Broadest ecosystem and integrations | Long-horizon coding and writing quality | Very low cost per token with open weights |
Our evaluation: on price per token and open-weight access, GLM is one of the strongest options available, and the coding plan undercuts most agent subscriptions. On ecosystem maturity and enterprise procurement, OpenAI and Anthropic still lead, and the Entity List status adds a compliance question for regulated buyers that no benchmark answers. For individual developers and small teams the trade-off is usually worth it; for regulated enterprises it is a decision to make deliberately.
🎯 Who Should Use Z.ai?
- Developers already paying for an agentic coding tool who want GLM-5.3 inside Claude Code or Cline for $18 a month.
- Cost-sensitive startups running high-volume inference, where $0.15 per million input tokens on GLM-5.3-Flash changes the unit economics.
- Privacy-conscious teams able to self-host, since the models ship under the MIT licence.
- US-based developers, because peak pricing runs 14:00-18:00 Singapore time, which is 02:00-06:00 in New York.
- Builders needing vision, OCR, speech or video without adding a second vendor, since one API covers all of them.
- OpenAI SDK users who want to test an alternative by changing a base URL rather than rewriting an integration.
🚫 Who Should Skip It
- Buyers who need refundable subscriptions. Coding plan payments are non-refundable, including annual plans.
- Teams locked to an unsupported harness. Plan benefits exclude other tools, and repeated violations risk restrictions.
- Regulated enterprises with US procurement constraints. The Entity List listing since January 2025 is a material consideration.
- Anyone expecting a simple per-request quota. Credits depend on input, cached input, output and MCP multipliers.
- Shoppers who want zero capacity surprises. February 2026 showed that compute availability is not guaranteed.
- Users who only need light chat. The free app and free Flash models already cover that without a plan.
❓ Frequently Asked Questions
Is there a free tier at Z.ai?
Yes. GLM-4.7-Flash and GLM-4.5-Flash are listed at zero cost for input and output tokens, and the consumer chat product is free to use, so you can evaluate the models before subscribing.
Can I get a refund on the GLM Coding Plan?
No. The policy states that subscriptions are non-refundable once purchased, even if the quota is unused, and that cancellations must be made at least 24 hours before the next billing date to stop auto-renewal.
Which coding tools does the plan support?
Claude Code, Cline, OpenCode, Roo Code, Kilo Code, Goose, Crush, Cursor, Windsurf and Trae. All supported tools share one credit pool, and usage elsewhere can lead to restrictions on plan benefits.
Is the API compatible with OpenAI libraries?
Yes. The endpoint is OpenAI-compatible at https://api.z.ai/api/paas/v4, so most integrations work after changing the base URL and key. Official Python and Java SDKs are also published, with function calling and context caching documented.
How does off-peak pricing work?
Model usage is charged at 50% of the standard credit rate outside peak hours. Peak is Monday to Friday, 14:00 to 18:00 Singapore time, which is 06:00 to 10:00 UTC, so New York developers are asleep for all of it while European late mornings overlap.
Is my data used for training?
Team plan documentation states that code, prompts and conversations are excluded from model training by default. For the individual plan, check the current terms in the console before sending sensitive material.
🏁 Final Verdict
Z.ai earns its place on the shortlist because it competes on fundamentals rather than marketing. GLM-5.3 at $1.40 and $4.40 per million tokens, GLM-5.3-Flash at $0.15 and $0.50, free Flash models, MIT-licensed weights and an $18-a-month coding plan for the tools developers already use is a combination few rivals match. The company behind it is listed, with a compute crunch in its recent past and a US Entity List entry, and its subscription terms are strict about refunds and supported tools.
Our recommendation: start free, prove the models on your own workload, then choose the plan tier that matches your agent usage rather than the biggest one available. If your workload is cost-sensitive, open-weight friendly and not subject to US procurement restrictions, Z.ai is one of the best value propositions in AI today. If any of those conditions fails, the cheaper option can become the expensive one.
Put GLM-5.3 Inside Your Coding Tool
Coding plans start at $18 a month with Vision, Web Search, Web Reader and Zread MCP included on every tier.
Disclosure: This article contains affiliate links. We may earn a commission if you buy through our links, at no extra cost to you. It does not influence our evaluation, and all prices, quotas and policy details above were taken from Z.ai and Z.AI public documentation at the time of writing.
Read next
Travel Hacks Guide Editorial Standards & Testing Methodology
At Travel Hacks Guide, we independently evaluate car rentals, international eSIMs, flight booking engines, and travel insurance. We never accept payment for ranking or recommendations. When you book via links on our site, we may earn an affiliate commission at no extra cost to you.