GLM 5.2
Z.ai's large general model, included in NeverNo's paid plans: a 200,000-token context, reasoning you switch on when needed, and the best score on our test of 12 sensitive programming requests (November 7, 2026).
Updated November 8, 2026
Readable on your devices only.
Your conversation is readable on your device, and on your other devices if you sync them. Nowhere else.
Unreadable to us.
Your message passes through without being written down; your chats are only ever stored encrypted.
The model doesn’t know you.
It gets the question, never your email, your account or your IP address.
Nothing is left along the way.
The reply comes back and is stored here, encrypted; even when you sync, your key never reaches us in the clear.
At a glance
| Maker | Z.ai |
|---|---|
| Context | 200,000 tokens |
| Longest reply | 8,192 tokens |
| Strengths in the app | Hard questions |
| Reasoning | On demand |
| Who can use it | Paid plans |
| How it’s paid | Included in the plan’s daily allowance, then credits |
| Privacy | Private: an open model run by NeverNo on providers chosen because they don’t keep requests. Your history stays encrypted in your browser. |
Price
What the app charges, read from the same table the servers bill with. 1 credit = $0.01.
| Typical message (2,000 tokens read, 600 written) | 1.44 credits ($0.0144) |
|---|---|
| Million tokens read | 369.25 credits ($3.69) |
| Million tokens written | 1,160.5 credits ($11.61) |
Included in the plan’s daily allowance first; these prices apply beyond it.
Measured
Our own measurements and one third-party score, each with its conditions and date. Figures are not rounded.
| Median time to the first word | 2.9 sOur client, 12 conversations of 3 turns, November 4, 2026 |
|---|---|
| Refused turns on our private test of sensitive requests | 5.6 %Same 12 conversations, November 4, 2026 |
| Legitimate programming requests that cautious models often refuse | 0 refused of 12, score 5.8/10Each model alone, scored blind by a judge of another family, November 7, 2026 |
| arena.ai Text Arena, overall score | 1,475Third-party leaderboard, page dated October 27, 2026 |
What it is
GLM 5.2 is the full-size model of Z.ai's GLM family, the bigger sibling of the free GLM 5.3 Flash. Our catalog files it under “smart”: it is a generalist meant for substantial work, from writing and analysis to software, rather than for the quickest possible reply. It is included in every paid plan, inside the daily message allowance, without credits.
Its standout result comes from our own test of sensitive programming requests. On November 7, 2026 we gave each model alone 12 legitimate programming requests of the kind cautious models often refuse, and a judge from another model family scored the answers blind. GLM 5.2 refused none of them and scored 5.8 out of 10, ahead of DeepSeek V4 Pro (4.2) and DeepSeek V4 Flash (4.1) in the same run. On the arena.ai Text Arena, its overall score was 1475 on the page dated October 27, 2026.
Reasoning is yours to steer: leave the default and it answers with the least reasoning it allows, switch reasoning off for brisk exchanges, or push it to its own full reasoning when a question deserves the extra time. The context window holds 200,000 tokens, and a reply can reach 8,192 tokens.
Like every open model on NeverNo, it runs privately, on providers chosen because they do not keep requests, with no identity attached; your conversation history is encrypted and kept in your browser.
Strengths
The best result we recorded on legitimate programming requests that cautious models often turn down: 0 refusals out of 12 and a blind score of 5.8 out of 10 on November 7, 2026.
A large working memory: 200,000 tokens of context, enough for a sizeable codebase excerpt, a long contract or a book-length manuscript discussed chapter by chapter.
Reasoning on demand: off for quick turns, on for hard ones, which lets you trade speed for depth message by message instead of switching models.
Included in paid plans without credits, so long sessions of heavy work do not draw down a balance.
Versatile beyond software: as a generalist, it suits research notes, structured essays, negotiation prep and planning as well, so one model can carry a whole working day from morning brief to evening wrap-up.
Limits
It takes longer to start than the DeepSeek models: a median of 2.9 seconds to the first word in our November 4, 2026 test of 12 three-turn conversations, versus 1.2 seconds for DeepSeek V4 Pro.
It refused 5.6% of turns on our private test of sensitive requests, twice the 2.8% measured for both DeepSeek V4 models. If a frank answer to a delicate question is what you need, those, or NeverNo Uncensored, are better bets.
It does not read images. For pictures, use GLM 5.3 Flash, which is free, or Kimi K3 on paid plans.
Replies stop at 8,192 tokens. For an entire long file in one answer, DeepSeek V4 Pro can write twice as much.
When to choose it
Code other assistants hesitate to write
Writing a fuzzer for your own parser, studying how a suspicious script behaves from its source, building a scraper, reverse-engineering a file format: legitimate engineering that it handled without refusals in our test.
Working across a large codebase
Paste several modules, the schema and the failing test into its 200,000-token context and ask for a refactor plan, a migration path or a review of how the pieces fit together.
Long-form writing and analysis
Structuring a report, critiquing a business plan, comparing two versions of a policy, keeping a novel's continuity across many chapters: tasks that reward a model able to hold a lot of material at once and notice when chapter twelve contradicts chapter three.
Prompt tips
Switch reasoning on for design questions, tricky bugs and anything with constraints that interact; leave it off for edits, renames and quick lookups, where it only adds waiting time.
Give it the surrounding code, not just the failing line. With 200,000 tokens available, context is cheap and it reduces invented function names and wrong assumptions about your project's conventions, folder layout or framework version.
For large outputs, ask for a plan first, then one part per message. Each reply is capped at 8,192 tokens, and splitting the work keeps every piece complete.
Frequently asked questions
Is GLM 5.2 free on NeverNo?
It is included in every paid plan, within the daily message allowance and without credits. On the Free plan, its sibling GLM 5.3 Flash is available instead.
Is GLM 5.2 private?
Yes. NeverNo runs it as a private model, on providers selected because they do not keep requests and without your name. Your chat history is stored encrypted in your own browser.
How good is GLM 5.2 at coding?
In our November 7, 2026 test of 12 legitimate programming requests that cautious models often refuse, it refused none and scored 5.8 out of 10, judged blind by a model from another family. It was the top score of that run.
GLM 5.2 or DeepSeek V4 Pro?
GLM 5.2 did better on our coding test and lets you switch reasoning on or off. DeepSeek V4 Pro always reasons, starts faster (1.2 seconds versus 2.9 in our November 4 test), refuses less often on sensitive requests (2.8% versus 5.6%) and writes replies up to 16,384 tokens.
What is GLM 5.2's context window?
200,000 tokens on NeverNo, with replies of up to 8,192 tokens.