Z.ai Review: AI Chatbot, Pricing & My Test Results

My rating
Accurate on facts and code. Two answers came back in Chinese.
Best for: facts, maths and code. Not for: images and audio.

Tested and written by Miles Foster
Based on 10 graded tests and my verdict. How I rate
I tested Z.ai hands-on with the same battery I give every AI chatbot: fact questions, reasoning problems, coding tasks, writing briefs, images and files. I used the free plan on its default model, GLM-5.3-Flash. Z.ai ranks #10 of the ten on my best AI chatbots page.
Every count and answer is in the Z.ai test results.
My Z.ai AI Chatbot Test Results
| Category | Grade | In short |
|---|---|---|
| Answer Accuracy & HallucinationEvery fact right, nothing invented | 5/5Excellent | |
| Reasoning & MathsSolved every problem it was given | 5/5Excellent | |
| CodingAll code ran and passed first time | 5/5Excellent | |
| Following Instructions & WritingMet nearly every constraint, one in Chinese | 5/5Excellent | |
| Context & Long-Document MemoryNot tested: 3 by default | 3/5Average | |
| Web Access & CitationsNot tested: 3 by default | 3/5Average | |
| Images, Files & VoiceRead some files, then stalled | 3/5Average | |
| Safety & RefusalsNot tested: 3 by default | 3/5Average | |
| Response SpeedSteady, but slow to start | 3/5Average | |
| Pricing & ValueNot tested: 3 by default | 3/5Average | |
| My verdict | +0.2 | |
| Overall | 4.0/5Good |
What Is Z.ai?
Z.ai is an AI chatbot web app at chat.z.ai. You ask questions, attach files and get answers, and some answers start a web search or run code in a sandbox.
The home screen offers starter tiles for Full-Stack, Writing, Data Insight, PPT and Web Site. On my free plan the model was GLM-5.3-Flash, and I left it unchanged.
How I Tested Z.ai
Every chatbot gets the same fixed battery, and I keep the first answer to every prompt. I started a fresh chat for each prompt and checked I was on the home page before I sent it.
My tools
Late in the run a notice said the maximum number of active sandboxes had been reached. I did not open the settings page to release one, so the requests after it could not finish.
Z.ai pros and cons
Pros
- Every captured fact answer right
- All 15 reasoning problems right
- All eight code answers passed
- Rejected the false premises captured
- Read a chart, a notice and a spreadsheet
- Met 48 of 50 writing constraints
Cons
- Two answers in Chinese to English prompts
- Image and audio requests stalled
- Sandbox notice stopped later requests
- A web search answer took 151 seconds
- About four seconds to first words
Z.ai Pricing
- Pricing & Value
- 3/5Average
My Z.ai account was on a free plan, with no plan name shown, running GLM-5.3-Flash by default. No message limit appeared in chat.
Z.ai plans and prices
Live data· last check 12 November 2026| Plan | Price a month | |
|---|---|---|
| FreeGLM-5.3-Flash by default; some answers use web search or a code sandbox | $0 | GLM-5.3-Flash by default; some answers use web search or a code sandbox |
- Struck-through prices of $18, $80 and $168 per month are shown next to the listed prices; the page does not say what they are.
- Monthly and Quarterly (20% off) billing options exist but their prices are not printed.
- Invite friends to earn up to 20% back in credits; Team tab has no prices shown.
A notice about the maximum number of active sandboxes appeared late in the run. Prices change, so check the plan page before you pay.
Z.ai Accuracy and Invented Facts
- Answer Accuracy & Hallucination
- 5/5Excellent
Z.ai answered all 29 fact questions I captured correctly and invented nothing. It rejected the three false-premise questions I captured, such as a third Nobel Prize that Marie Curie never won.
One answer started a web search and listed its sources, which helps you check it.
Z.ai Reasoning and Maths
- Reasoning & Maths
- 5/5Excellent
Z.ai solved all 15 reasoning and maths problems on its default model.
Fifteen problems is a small set, so you may want a harder set to separate the top chatbots.
Z.ai Coding
- Coding
- 5/5Excellent
Z.ai wrote code that passed all eight of my checks the first time, with no fixes from me, so you can paste it and run it.
Two tasks returned no code text for me to run. One explanation under a code answer came in Chinese.
Z.ai Following Instructions
- Following Instructions & Writing
- 5/5Excellent
Z.ai met 48 of 50 constraints. A three-sentence product description came wrapped in extra sentences, and an email ended with a note about its word count after the sign-off.
It handled the contradictory brief well, by saying the two lengths cannot both hold and offering both. But it wrote that answer in Chinese, to an English prompt, so you may need a translator.
Z.ai Images and Files
- Images, Files & Voice
- 3/5Average
Z.ai read five images and files correctly: a photo, a chart, a notice, a maths problem and a spreadsheet. It gave the spreadsheet's North revenue and the region with the most units.
The image request and the audio request stayed on "Ran 1 commands" with no result. The handwriting and PDF requests returned no answer text, and the table image hit the sandbox notice.
Z.ai Response Speed
- Response Speed
- 3/5Average
Z.ai took about four seconds to show its first words and about eight to finish. The slowest answer started a web search and took 151 seconds.
The next slowest took 27 seconds. I timed with a script, so treat the figures as close, not exact.
Final verdict: is Z.ai worth it?
Z.ai is worth trying if you want accurate answers, maths and code from a free chatbot. Everything I could count on facts, reasoning and code was right.
It is a weaker fit if you need images, audio or file answers you can count on, or an answer always in English. Image and audio requests stalled, and two answers came in Chinese.
The ten tests average 3.8 out of 5. The free plan let me run 6 of 10 tests, so I added 0.2 and rate it 4.0.
See the Z.ai test results and how I rate.
Best for: People who want accurate answers, maths and code from a chatbot and do not mind an occasional answer in Chinese.
Video review · 2:40
Watch me test Z.ai
The whole test in about three minutes: facts, reasoning, code, writing briefs, images, files and speed. Two answers came in Chinese. Captions and the full transcript are one click away, and the video has a page of its own.
Transcript
Hi, I'm Miles. I tested Z. ai, a free chatbot, with the same fixed battery I give every chatbot. Here's the short version. It got every fact, reasoning problem and code answer right. But two answers came back in Chinese, and its image and audio requests stalled. My rating is four out of five.
So what is Z. ai? It's a chatbot web app. You ask questions, attach files and get answers, and some answers start a web search or run code in a sandbox. On my free plan the model was GLM five point three Flash, and I left it unchanged.
Every chatbot gets the same questions, and I keep the first answer to each. I started a fresh chat for every prompt. Late in the run a notice said the maximum number of active sandboxes had been reached, so the requests after it could not finish.
On facts, Z. ai got every answer I captured right and invented nothing. It rejected the three false premise questions I captured, like a third Nobel Prize that Marie Curie never won.
It solved all fifteen reasoning and maths problems on its default model. Fifteen is a small set, so a harder set could tell the top chatbots apart.
All eight code answers passed first time, with no fixes from me. Two tasks returned no code to run, and one explanation under a code answer came in Chinese.
On writing briefs, Z. ai met forty-eight of fifty constraints. A three sentence description came wrapped in extra sentences. It handled the contradictory brief well, but answered in Chinese to an English prompt.
It read a photo, a chart, a notice, a maths problem and a spreadsheet correctly. The image request and the audio request stayed on a running line with no result. The handwriting and PDF requests returned no answer text.
Z. ai took about four seconds to show its first words and about eight to finish. The slowest answer started a web search and took a hundred and fifty-one seconds.
My account was on a free plan with no plan name shown, running the default model. No message limit appeared in chat, but the sandbox notice did, late in the run.
So, is it worth trying? Yes, for accurate answers, maths and code. It is weaker for images, audio, or answers that must stay in English. The ten tests average three point eight. The free plan let me run six of ten tests, so I added point two. My rating is four out of five.
Chapters · 13
Frequently asked questions
Is Z.ai free?
Yes, I used a free plan. No plan name was shown, and I bought nothing.
Is Z.ai accurate?
Yes, on what I ran. It got 29 of 29 captured fact answers and 15 of 15 reasoning problems right.
Which model did I test?
GLM-5.3-Flash. It was the default on my free plan, and I left it unchanged.
Does Z.ai answer in English?
Mostly. Two answers came in Chinese to English prompts: the contradictory brief and the explanation under one code answer.
Can Z.ai make images or read audio?
Not for me. My image request and my audio request stayed on "Ran 1 commands" with no result.
Miles FosterOwner and tester, RankDudeMiles Foster is a pen name: the owner of RankDude, a private person, runs every test himself.
Disclosure. I tested on the free demo account anyone can open on Z.ai’s site. I signed up myself and paid Z.ai nothing. Z.ai didn’t see this review first and had no say in my rating. How RankDude makes money is on the disclosure page.