Z.ai and the GLM infrastructure family
Z.ai is the Chinese company behind the GLM family, and its pricing table today spans fourteen text models and six vision models, three of which carry zero cost. The flagship unit is GLM-5.2: a one-million-token operating window, an output ceiling of up to 128K tokens, at a rate of $1.4 input and $4.4 output.
- three zero-cost models on the pricing table
- coding plan from $18
- GLM-5.2 at a one-million-token window
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. Change log
Model families
The GLM-5 production line, whose three consecutive versions all remain active on the pricing table.
Versions
| Versions | Deployed | Status | Model families |
|---|---|---|---|
| GLM-4.6V | current | GLM-4 | |
| GLM-4.7-FlashX | current | GLM-4 | |
| GLM-4.7 | current | GLM-4 | |
| GLM-5.1 | current | GLM-5 | |
| GLM-5.2 Max | current | GLM-5 | |
| GLM-5.3-Flash | current | GLM-5 | |
| GLM-5.3 | current | GLM-5 | |
| GLM-OCR | current | GLM-4 |
Three zero-dollar rows, and none of them is the new generation
The Z.ai pricing table logs three models at zero cost: GLM-4.7-Flash, GLM-4.5-Flash, and the vision model GLM-4.6V-Flash. No other vendor in this group carries a row like that in its own specification, and for anyone provisioning a test without a foreign card, this is the one zero-cost entry point.
The boundary on this offer needs documenting as well. All three sit in generation 4, not generation 5. The model that earns points in our coding table is GLM-5.2, and that unit is not free. Free here provisions a trial, not a production stack.
A plan that allocates credits instead of tokens
Alongside the API, Z.ai runs a coding plan starting at $18 a month across three tiers. Its accounting model differs from the API: it allocates credits and enforces two ceilings simultaneously, one over five hours and one over the week, and whichever fills first is the one applied. The entry tier carries 2,000 five-hour credits and 10,000 weekly credits. Off-peak usage is billed at half rate. The documentation logs the combined saving at up to 92 percent against API pricing.
This is the same subscription architecture familiar to anyone running Claude Code, and the plan is compatible with Claude Code, Cline, and OpenCode. If your operational tooling is one of those three, migration is a matter of swapping the key and the endpoint.
Where this vendor is genuinely transparent, and where it is not
Z.ai documentation is the one source that explicitly documents the reasoning_effort parameter with the value max. What looks like a model name on the Arena leaderboard is, in this specification, a parameter setting. That is why every page in this reference that needs to substantiate that claim links back to this vendor.
Against that, none of the model guide records carry a release date. It cannot be extracted from Z.ai own records whether GLM-5 shipped months ago or GLM-5.2 shipped weeks ago. For a reference where the entire structure is built on dating, that gap is a serious documentation shortfall.
Documented strengths
- The only vendor in this group with zero-cost models on its official pricing table: GLM-4.7-Flash, GLM-4.5-Flash, and GLM-4.6V-Flash
- A coding plan from $18 a month compatible with Claude Code, Cline, and OpenCode
- A one-million-token operating window on GLM-5.2 with an output ceiling of up to 128K tokens
- A deep catalogue: fourteen text and six vision models with published pricing specifications, from $0.03 to $8.9
- Its documentation is the clearest source establishing that max is an effort-level parameter, not a separate model
Known weaknesses and limits
- No model guide record carries a release date, so the operational age of any version cannot be extracted from the vendor own sources.
- It publishes no geographic-coverage specification, so our Iran column stays unchecked.
- Every zero-cost model is generation 4, not 5, so what is free is not what scores in our table.
- GLM-5.1 carries the exact rate of GLM-5.2 while running a fifth of its context window. Staying on 5.1 has no cost justification.
- The catalogue carries no image or video generation unit. Its vision models analyze images, they do not produce them.
Technical verdict
If you want to provision a zero-cost test, or run Claude Code against a fixed monthly budget, Z.ai is the most practical pick in this group. If you need to establish the operational age of your model, that data has to be extracted from a source other than this vendor.
Frequently raised questions, with documented answers
Is GLM-5.2 Max a separate model
No. Z.ai own documentation names the unit glm-5.2 and logs max as a value of the reasoning_effort parameter. The pricing and context-window specifications belong to glm-5.2 itself.
Is the free Z.ai model sufficient for real workloads
For a trial, yes; for production, unlikely. All three zero-cost rows on the pricing table are generation-4 models, while the unit that scores in our coding ranking is GLM-5.2, which carries no zero-dollar row.
What capacity does the coding plan actually allocate
The entry tier is $18 a month with 2,000 credits per five hours and 10,000 per week, and whichever ceiling fills first applies. Off-peak usage is billed at half rate. The pricing specifications for the two higher tiers are not published on that record.