GLM and Z.ai: what you get free and paid, and the right setup for every task
GLM is the model family from Z.ai (the international service of Zhipu AI), and Z.ai Chat is a free chatbot with Agent, slides, website building and Deep Research. What you pay for is mainly the GLM Coding Plan, which runs inside Claude Code, Cline, Cursor and similar tools. Here is what you really get, where GLM is weak and what to use instead. Checked October 2026.
01What GLM and Z.ai are and who they suit
GLM is the name of the models; Z.ai is the international service that runs them. The free chat at chat.z.ai uses GLM-5.3-Flash by default: a model with a one-million-token context window that also understands images and video. The stronger GLM-5.3 is available through the Coding Plan and the API. Both are open weights: you can download them from Hugging Face and run them yourself, if you have serious GPU hardware.
The chat suits anyone who wants a strong tool without paying: a deck that exports to PPTX, a small website that goes live, research with sources, and summaries of long documents. The Coding Plan suits an engineer or office already using Claude Code, Cline or Cursor who wants to run them on GLM at a low price.
Less suitable: if you need polished Hebrew, image and video generation in the chat, a link to Outlook or Drive, or a phone app. And if you work with US government or defence clients, read the privacy card below first.
02The Coding Plan as a fuel card
The Z.ai chat is a free rental car: it drives well, but you do not choose the model and you do not know when the tank runs dry. The Coding Plan is a cheap fuel card for a car you already own, such as Claude Code or Cursor. It works only at listed stations, meaning the supported coding tools, and not for the company fleet, meaning the apps and bots you build yourself. Those need the separate API.
03Free vs paid: what you really get
Z.ai Chat, free
- Chat on GLM-5.3-Flash with a one-million-token context window; other GLM models selectable.
- Agent mode for multi-step tasks.
- AI Slides: a finished deck exported to PDF, PPTX or images.
- Full Stack: builds a website or app from a prompt and can deploy it.
- Deep Research with web search and cited sources.
- File upload plus image and video understanding: tables from images, charts, screenshots.
- Temporary chat, history deletion and chat sharing.
Limit: No published numeric limits: the Agent quota and the daily slide-export limit show up only when you hit them. No Hebrew interface, and individual users' content may be used for training until you switch off Improve Model.
GLM Coding Plan (for coding tools)
Lite: 18 USD a month; 14.4 USD a month billed quarterly; 12.6 USD a month billed yearly. Pro: 80 USD a month (56 USD billed yearly). Max: 168 USD a month (117.6 USD billed yearly). Team: 88 USD per seat a month (minimum 2 seats).
- GLM-5.3 and GLM-5.3-Flash inside Claude Code, Cline, Kilo Code, Roo Code, Cursor, Codex, OpenCode and ZCode, over 20 tools.
- Large credit allowances: on Lite, 2,000 per 5 hours and 10,000 per week.
- Off-peak and weekend use costs half the credits. Peak is only Mon–Fri 14:00–18:00 Singapore time: 09:00–13:00 Israel time until 25 October 2026, and 08:00–12:00 after the switch to winter time.
- Vision, Web Search, Web Reader and Zread MCP servers included.
- One-command setup with the Coding Tool Helper.
- AutoClaw and ZCode, Z.ai's desktop apps, work with the plan.
When paying is worth it: When you already use a coding agent such as Claude Code or Cline and want to make it much cheaper. We found no separate paid plan for the chat itself: the chat's Upgrade button leads to the Coding Plan, and it is not documented that this raises chat limits.
04The weak sides
In the free version
- No image generation, video generation or voice chat clearly offered in the chat; they exist only as paid API models.
- No connectors to Gmail, Outlook, Google Drive or OneDrive, and no Office add-ins.
- Web only: we found no official mobile or desktop chat app.
- Interface in English and Chinese only, and Hebrew quality is not documented.
- Limits are not published; the Agent quota and slide-export limit appear only when you reach them.
- Individual users' content may be used for training; you must switch off Improve Model yourself.
Even when paying
- The Coding Plan works only in supported tools: no use as an API for your own website, bot or app, and no account sharing.
- Non-refundable and auto-renewing; the subscription terms say cancel 24 hours before renewal, the usage policy says 3 days.
- Usage is measured in abstract credits, and peak-hour use counts double.
- Credit cards that require 3DS verification are not supported, and many Israeli and European cards use it.
- Individual plans carry no contractual no-training promise like the Team Plan has.
- We found no paid chat plan, so the chat's weak sides stay even when you pay.
05Plans and prices
| Plan | Price per month | For whom | Main limits |
|---|---|---|---|
| Z.ai Chat (Free) | 0 USD | Anyone who wants a strong free chatbot | Limits not published; Agent quota; daily slide-export limit; web only |
| Coding Plan Lite | 18 USD; 14.4 USD quarterly; 12.6 USD yearly | An engineer or student using Claude Code, Cline or ZCode | 2,000 credits per 5 hours, 10,000 per week; supported tools only; personal use |
| Coding Plan Pro | 80 USD; 64 USD quarterly; 56 USD yearly | Heavy daily use on mid-sized projects | 12,000 credits per 5 hours, 60,000 per week |
| Coding Plan Max | 168 USD; 134.4 USD quarterly; 117.6 USD yearly | Several agent projects in parallel | 28,000 credits per 5 hours, 140,000 per week |
| Team Standard | 88 USD per seat; yearly, the plan page shows from 79.2 USD a month (checkout price not verified) | Small offices that want admin and central billing | Minimum 2 seats; 15,000 credits per 5 hours, 66,000 per week per seat |
| Team Premium | 188 USD per seat; yearly, the plan page shows from 169.2 USD a month (checkout price not verified) | Team members running agents all day | Minimum 2 seats; 35,000 credits per 5 hours, 155,000 per week per seat; cannot be mixed with Standard |
| Z.ai API | Pay per use: GLM-5.3 1.4 USD input and 4.4 USD output per million tokens; some Flash models free | Developers building their own tools and bots | Prepaid balance; billing shown with a one-day delay |
| Full comparison | AI Price Comparison: Claude, ChatGPT, Gemini, Copilot, DeepSeek, Qwen, Kimi and GLM | All assistants in one table | Prices in USD; local Israeli price and VAT not shown |
06The best setup for each task
Write an email or letterA clear, businesslike email, mainly in English; in Hebrew, read and fix before sending.4
Free way
- Z.ai Chat (GLM-5.3-Flash)A good free writer; Hebrew is not officially supported, so check the wording.
- Temporary chatFor a sensitive letter: a chat that is not kept in history.
Paid way
- The same chat (we found no paid chat tier)No email connector, so copy and paste into Outlook or Gmail.
- A partner for Hebrew letters from your inboxIf you want drafting from the real thread, see Claude for the office or ChatGPT guide.
Write a polite but firm English email to a supplier about a delayed delivery of anchors: 150 words, a subject line and three action items with dates.
Write a technical reportA report with fixed chapters, data from your files and flagged assumptions.4
Free way
- Z.ai Chat with file uploadA one-million-token model reads long notes and site photos.
- Agent modeWorks on the report in several steps and can produce a document, within the Agent quota.
Paid way
- AutoClaw on the Coding PlanA desktop work assistant with workflows and knowledge bases.
- Vision MCP ServerUnderstands photos, screenshots and diagrams inside your agent tool.
Upload notes and photos, then: "Write a report with Scope, Findings, Calculations, Recommendations; mark every assumption [ASSUMPTION] and list missing data at the end."
Summarise long documents and PDFsA short summary followed by a table of every numeric requirement with page numbers.4
Free way
- File upload in Z.ai ChatGLM-5.3-Flash reads very long documents thanks to its one-million-token window.
- Image understandingReads tables and charts even when they are images inside the document.
Paid way
- GLM-OCR via the APILayout parsing for scanned PDFs at 0.03 USD per million tokens.
- Vision MCP ServerReads scans and diagrams inside Claude Code or Cline.
"Summarise this PDF in 10 bullets, then a table of every numeric requirement with its page number."
Translate documentsGood translation for English, Chinese and European languages; Hebrew needs checking or another tool.4
Free way
- Z.ai ChatGood for English, Chinese and European languages; Hebrew is not in the official language lists.
- Google Translate for HebrewFree and quick for Hebrew texts, since Hebrew is not in GLM's official language lists.
Paid way
- Translation Agent (API)40 languages with glossaries, but Hebrew is not supported.
- An assistant with official Hebrew supportFor important Hebrew documents see Claude for the office, ChatGPT guide or Gemini guide.
Give a glossary first: "Translate to German keeping these terms: anchor=Dübel, base plate=Ankerplatte; keep units and numbers unchanged."
Solve an engineering problem with a checked calculationA step-by-step solution with units, and code that recomputes the numbers.5
Free way
- Z.ai Chat (thinking model)Strong reasoning; ask for a detailed solution with units.
- Agent mode to run codeAsk it to write and run code that checks the numbers.
- ADIT's AAS and ARS calculatorsFor anchor and rebar calculations: a free, independent check.
Paid way
- Claude Code with the GLM Coding PlanWrites a Python script that computes and verifies the result on your computer.
- Cline in VS CodeThe same idea inside VS Code, with GLM-5.3.
"Solve step by step with units, then write a Python script that recomputes every number and compare; flag differences." Check the clauses against the standard yourself.
Analyse Excel data and make chartsA clean summary table, a chart and the code used, so you can check it.4
Free way
- Agent mode with file uploadUpload a CSV or XLSX and ask for charts; quality is not officially documented.
- Ask for the Python codeSo you can check every calculation and run it again.
Paid way
- AutoClaw on the Coding PlanZ.ai's Coding Plan docs list 'upload business data, get charts automatically' as an office use; AutoClaw is its desktop work app.
- Claude Code with GLMWorks directly on the data files in your folder.
"Load this CSV, clean empty rows, show monthly totals in a table and a bar chart, and give me the Python code used."
Create a presentationA designed deck exported to PPTX or PDF, ready for final edits.4
Free way
- AI Slides in Z.ai ChatOne of GLM's best free features; exports to PDF and PPTX.
- Final edits in PowerPointOpen the PPTX and fix details; there is a daily export limit, so export when the deck is nearly done.
Paid way
- Slide/Poster Agent (API)Slides and posters through code, 0.7 USD per million tokens, PDF export (beta).
- The same AI SlidesIt is not documented that paying raises the chat's daily export limit.
"Create a 10-slide deck for a client meeting about chemical anchors: problem, solution, 3 case studies, cost table, next steps; clean corporate style in blue and grey."
Draw an image or illustrationGLM is weak here in the chat; a dedicated image tool is better.4
Free way
- Microsoft Designer or CopilotWe found no clearly offered image generation in the Z.ai chat; see Copilot guide.
- Z.ai Chat to write the promptDrafts a detailed prompt you paste into the image tool.
Paid way
- GLM-Image (API)0.015 USD per image, with readable text inside the image.
- CogView-4 (API)Another image model, 0.01 USD per image.
For GLM-Image: "Clean isometric illustration of a concrete wall with three anchors, labels in English, white background."
Technical drawings, diagrams and flowchartsMermaid or SVG code you paste into a viewer; there is no CAD here.4
Free way
- Z.ai Chat for Mermaid or SVG codeGenerates diagram code you paste into a viewer.
- draw.io (free)Open the code GLM produced and tidy it by hand.
Paid way
- draw.io with Mermaid or XML from GLMGLM has no drawing or CAD tool of its own, even when paying.
- Vision MCP ServerReads an existing diagram from a screenshot so you can redraw or improve it.
"Draw a Mermaid flowchart of our anchor-selection process: load type → base material → edge distance → product choice."
Video and animationGLM writes the script and storyboard; the video itself is made elsewhere or via the API.4
Free way
- Clipchamp or Google VidsVideo generation is not offered in the free chat; Clipchamp is a free online video editor.
- Z.ai Chat for the scriptWrites a storyboard and voice-over text for the video.
Paid way
- CogVideoX-3 (API)0.2 USD per video, from text or an image.
- GLM-Image (API)Images for the video's frames and slides.
"Write a 30-second storyboard, 6 shots, for a product video on fixing a railing."
Web research with sourcesAn answer where every claim comes with a link and a date.4
Free way
- Deep Research in Z.ai ChatWeb search on by default, with cited sources.
- Normal search in the chatShows how many keywords and sources were used for the answer.
Paid way
- Web Search MCP ServerWeb search inside your agent tool.
- Web Reader MCP ServerFetches and parses a full web page.
"Research X; list every claim with its URL and date; separate official sources from blogs."
Search standards and regulationsAnswers from the standard you bought, with clause numbers; without it, only general direction.4
Free way
- Deep Research (with caution)Finds public pages but has no access to paid standards; verify every clause.
- Upload the standard you ownAnswers only from the PDF you uploaded.
Paid way
- The Standards Institution of Israel (SII)GLM has no standards-database link; buy the standard from the official body and upload it to the chat.
- Web Reader MCP ServerReads full public regulation pages inside your agent tool.
"Answer only from this PDF; quote clause numbers; say NOT FOUND if absent."
Write code or a small engineering toolGLM's strong area: a working tool with tests, inside the coding agent you already know.6
Free way
- Z.ai Chat (code and Full Stack)GLM is coding-focused, and the chat is free.
- Open weights on Hugging FaceRun GLM-5.3-Flash yourself (MIT licence); needs serious GPU hardware.
Paid way
- Claude Code with GLMGLM-5.3 inside Claude Code, from 18 USD a month.
- Cline or Kilo Code in VS CodeVS Code extensions with an official Coding Plan setup.
- CursorOfficial guide to set GLM as a custom model.
- Coding Tool HelperA wizard that installs and configures the tool and the MCP servers in one command.
Run npx @z_ai/coding-helper to connect Claude Code to GLM, then: "Build a single-file HTML calculator for beam deflection with input validation and unit tests."
Build a website or landing pageA working site that goes live, plus files you can download.5
Free way
- Full Stack in Z.ai ChatBuilds a site from a prompt and deploys it, within the Agent quota.
- Ask for downloadable filesSo the site does not depend on Z.ai's hosting.
Paid way
- ZCodeZ.ai's development environment for long coding tasks; full control of the code.
- Claude Code with the Coding PlanBuilds the site in your folder; you deploy to your own host.
- CursorA code editor with GLM as the model.
"Build a one-page landing site for a small engineering office: hero, services, 3 projects, contact form; mobile-first; give me the files to download."
Organise files and folders on the PCThe chat cannot touch your PC; it writes a script you review and run.4
Free way
- Z.ai Chat for a PowerShell scriptThe web chat cannot reach your computer; ask for a script and review it before running.
- Dry-run modeAsk the script to only print what it would do before it moves anything.
Paid way
- AutoClawA desktop agent that works on folders you open to it.
- Claude Code with the Coding PlanA terminal agent that acts on the folder it runs in.
"Write a PowerShell script that sorts files in Downloads into folders by year and type, with a dry-run mode that only prints the moves."
Meetings: transcription and minutesNo meeting recorder; paste a transcript and get tidy minutes.4
Free way
- Transcript from Teams or ZoomGLM has no meeting recorder; download the transcript from your meeting tool.
- Z.ai Chat for the minutesTurns a transcript into a summary with decisions and tasks.
Paid way
- GLM-ASR-2512 (API)Speech-to-text at about 0.0024 USD per minute; Hebrew is not among the named languages.
- A tool with Hebrew supportFor meetings in Hebrew see Copilot guide or Gemini guide.
"From this transcript write minutes: decisions, action items (owner, due date), open questions."
Customer service and repeated repliesConsistent replies that follow your rules and never promise what you cannot.4
Free way
- Z.ai Chat with a saved instruction textWe found no custom-bot builder; paste your reply rules each time.
- A standing rules fileKeep the rules and sample replies in one file and paste it before each task.
Paid way
- Z.ai API with GLM-5.3-FlashFor your own reply bot, 0.15 USD input and 0.50 USD output per million tokens. Not the Coding Plan: bots are prohibited there.
- Free Flash models on the APIGLM-4.7-Flash and GLM-4.5-Flash are free, for a first trial.
"You answer customer emails for an anchor supplier. Rules: … Draft a reply to this email; never promise delivery dates."
Learn a new subject (tutor)Short lessons with check questions, and the free chat is enough here.4
Free way
- Z.ai ChatExplains step by step and asks check questions.
- Deep Research for sourcesFinds background material with links.
Paid way
- No gain from paying hereWe found no dedicated study mode; the free chat already covers tutoring.
- Zread MCP ServerFor learning code: reads and explains a whole GitHub repository.
"Teach me reinforced-concrete punching shear in 5 short lessons; after each, give 3 quiz questions and wait for my answers."
Marketing and SEO contentA well-structured article with headings and an FAQ; proofread Hebrew before publishing.4
Free way
- Z.ai Chat with Deep ResearchDrafts articles and researches competitors; Hebrew quality is not documented.
- AI Slides for sales materialA short deck or product sheet as PDF from the same content.
Paid way
- Web Search MCP ServerBrings current data from the web into the draft.
- Web Reader MCP ServerReads live pages, such as competitors', into your agent tool.
"Write a 900-word article on choosing chemical anchors, with H2 headings, an FAQ of 5 questions and a meta description under 155 characters."
Automate a repeating workflow (agents)A fixed process that runs end to end, started small and checked at each step.5
Free way
- Agent mode in Z.ai ChatMulti-step tasks in a sandbox, within the Agent quota.
- Full Stack for an internal toolBuilds a small web tool that handles the repeating part.
Paid way
- AutoClawAccording to Z.ai, runs complex workflows end to end.
- OpenClaw and other supported toolsThe list of official agent tools that work with the plan.
- Claude Code with GLMAn agent that runs the process on your files.
Start small: "Every time I give you a supplier price list, convert it to our CSV template, flag price changes over 5%, and draft an email summary."
07Privacy and Hebrew
Training on your data
Under the Terms of Service, Z.ai may use individual users' content to improve its services and models. The chat settings have an Improve Model switch; its default state is not published, so check that it is off. You can also delete single chats or the whole history.
Team and API
On the Team Plan, data is not used for model training by default. On the API, content is not used to improve the service unless you explicitly agree, and the DPA says it is processed in real time and not saved. The individual Coding Plan card says Default data privacy, without a detailed definition.
Hebrew, RTL and mobile
The interface is in English and Chinese only, and the model cards list just those two languages. The model reads and writes Hebrew, but Z.ai does not document the quality, so check every text. We found no official mobile or desktop chat app; store apps named Z.ai are not official.
Region, Singapore and the Entity List
The international service is operated from Singapore (JINGSHENG HENGXING TECHNOLOGY PTE.LTD), and data is generally processed in Singapore; it may also be transferred to affiliates in the Z.ai Group. Israel is not on the restricted-regions list. Since 16 January 2025, Zhipu's Beijing entities are on the US Entity List (the stated reason: advancing China's military modernization through AI): this restricts US exports to them and is not a ban on using Z.ai. If you work with US government or defence clients, check with compliance. Either way, do not paste confidential drawings, contracts or personal data into the free chat.
08Set it up in 10 minutes
- 1
Open the chat
Go to chat.z.ai in your browser and sign up for free. We found no official app, so do not install store apps named Z.ai.
- 2
Switch off training
In the chat settings, make sure Improve Model is off. For a sensitive question use Temporary chat.
- 3
Try the modes
Pick AI Slides, Deep Research or Full Stack and give each one small task, to see what each does.
- 4
A first real task
Upload a specification PDF and ask for a summary with every numeric requirement and its page number. Check three requirements against the source.
- 5
Only if you use a coding agent: the Coding Plan
First check that your card does not require 3DS. After buying, run npx @z_ai/coding-helper to connect Claude Code or another tool. Run heavy jobs outside Mon–Fri 09:00–13:00 Israel time (08:00–12:00 from 25 October 2026), or at the weekend.
- 6
Set a cancellation reminder
There are no refunds and the plan renews automatically. Set a reminder at least 3 days before renewal, because two official pages give different deadlines.
09FAQ
Is the free version enough for a small office?
For decks, summaries, research and English emails, yes. Missing are email and cloud-file links, images, a phone app and documented Hebrew. And on Free, switch off Improve Model before uploading office documents.
Does Z.ai have a paid chat plan like ChatGPT Plus?
We found none. The chat's Upgrade button leads to the GLM Coding Plan, and it is not documented that the plan raises chat limits. If you want a paid chat with Hebrew and connectors, look at ChatGPT guide or Claude for the office.
Can I use the Coding Plan for a bot on my website?
No. The Coding Plan is only for supported coding tools and personal use. For your own bot, website or app, use the Z.ai API and pay per use; GLM-5.3-Flash costs 0.15 USD input and 0.50 USD output per million tokens.
Is it allowed to use Z.ai given the US Entity List?
The listing (effective 16 January 2025) restricts exports of US technology to Zhipu entities. It does not forbid users from using Z.ai. Companies with US government or defence clients should check with compliance; another option is running the open weights on your own server.
Which plan for a team of 3 developers?
Team Standard: 88 USD per seat per month, minimum 2 seats, so 264 USD a month for three, with user management, central billing and no training by default. A personal Lite subscription may not be shared between employees.
Can I trust its engineering calculations?
Only after checking. Ask for a solution with units and a Python script that recomputes it, and check against the standard. For anchor and rebar calculations, ADIT's free AAS and ARS calculators (https://www.adit.org.il/all-the-software/) are a reliable independent check.