GLM-5.2 Max
GLM-5.2 scores 1588 on the WebDev Arena leaderboard and costs $1.4 in and $4.4 out, under a fifth of what Opus 5 charges for output. Its total in our coding table is only 2.3, and the section below explains why that number says more about our data than about the model.
- 1588 on WebDev Arena
- $4.40 per million output tokens
- 1M token context window
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Max is a parameter, not the name of a model
The Arena leaderboard shows a row called glm-5.2-max. Z.ai's own documentation calls that model glm-5.2 and lists max as one of the values of the reasoning_effort parameter. What earned the ranking is a setting applied to the model.
We make the same point on the Kimi page, but there we could only say that Moonshot documents no such model. Here the vendor shows you the parameter itself, which makes this the cleanest evidence for the claim: any comparison that presents GLM-5.2 Max as a separate product with its own price has mistaken a leaderboard row for a model catalogue.
Five times better value, and 1.2 points for it
Opus 5 charges $25 per million output tokens; GLM-5.2 charges $4.40. Divide the leaderboard rating by the output price and GLM lands on 360.9 against Opus 5's 67.7. On that ratio it is more than five times ahead.
In our score-per-dollar column it still takes only 1.2 of twenty points. DeepSeek V4 Flash is the reason: it sits at 5660 in the same column and stretches the scale far enough that the distance between GLM and Opus 5 is squashed flat at the bottom. Min-max normalisation behaves that way and we do not adjust it, but the column should not be read as saying GLM is anywhere near Opus 5 on value for money. It is not close. It is far ahead.
Two empty columns are what hold its rank down
GLM-5.2 totals 2.3 out of 100 in the coding table, and fifteen of those hundred points have no evidence at all behind them. One is the general text leaderboard: we have not read a GLM row there, so those ten points sit empty. The other is Iran access: Z.ai publishes no supported-countries list and we have not tested it from an Iranian connection.
Read 2.3 this way. In the one column where the evidence is complete, GLM scored 1588, which is 104 points below the top of the table and three above the bottom of it. The rest of the number comes from missing data rather than from measurement. Fill those two columns and the rank moves by itself, with nothing placed by hand.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- not checked
- Payment
- not checked
- Free tier
- no
How we checked: Z.ai publishes no supported-countries list. Until we measure it ourselves from an Iranian connection it stays unchecked.
What it is good at
- An output price of $4.40, under a fifth of Opus 5
- A one million token window with 128k output, level with the pricier models in the table
- The vendor documents the reasoning_effort parameter openly, so max-mode behaviour is reproducible
- More than five times ahead of Opus 5 on rating per output dollar
Where it falls short
- On the web-dev leaderboard it is 104 points below the top of the table and only three above the bottom.
- We have not read its row on the general text leaderboard, so ten points of our table stay empty for it.
- Z.ai publishes no supported-countries list, so the Iran column stays unchecked.
- It scores zero in the tooling column: no official command line tool and no editor extension in the docs.
- The 2.3 total reflects our missing data more than the model quality, which makes it useless for comparison against the leaders.
Our take
For high-volume work where token cost decides and top-tier quality is not required, GLM-5.2 is a serious option. But until we have read its text-leaderboard row and its Iran access, we will not recommend it to anyone about to build on one model: a choice whose specs are half unverified is not a choice.
Questions people actually ask
What does GLM-5.2 cost
One dollar forty per million input tokens and four dollars forty per million output tokens, from Z.ai's own pricing page.
What does reasoning_effort set to max mean
It tells the model to reason more before answering the same request. The model does not change, only the amount of work it does first, and the glm-5.2-max row on the leaderboard is that mode.
Why is GLM scored so low in your table
Because fifteen of a hundred points have no evidence behind them: its general text leaderboard row and the Iran access column. In its one complete column it scored 1588, which puts its real distance from the leader at 104 points.
Does GLM work from Iran
Unchecked. Z.ai publishes no supported-countries list and we have not tried it from an Iranian connection, so any firm answer today would be a guess.
Sources
- Z.ai GLM-5.2 guide vendor source 12 August 2026
- Z.ai pricing vendor source 12 August 2026