Arena.ai evaluated GLM-5.2 (Max) in three types of in-platform evaluations in June 2026, with the following conclusions:
GLM-5.2 (Max) ranked second overall, scoring 29 points higher than Claude Opus 4.7 (Thinking) and trailing only Claude Fable 5; it was the highest-ranked open-source model.
Sub-rankings: second in React and fourth in HTML; among open-source models, it had the largest lead over Kimi-K2.6 and MiniMax-M3.
Trajectory: GLM-series scores in Code Arena: Frontend rose from 1408 for GLM-4.6 to 1595 for GLM-5.2 (Max)—surpassing Claude Opus 4.8 and closing in on Claude Fable 5 (1665 points).
Update post on August 4: GLM-5.2 (Max) still ranked second overall in Frontend Code Arena (first in the open-weight group).
On June 18, GLM-5.2 (Max) entered the top 10 and was the strongest open-weight result measured at the time: confirmed task success increased by 9.4%, and the praise-complaint ratio increased by 14.9% (relative to the baseline).
On June 26, Arena published an analysis of Agent Arena token efficiency (the model can call search, filesystem, and terminal tools to complete complex workflows such as writing code, creating slides, conducting research, building applications, and analyzing documents).
GLM-5.2 (Max) ranked #25 overall, close to GLM-5.1; its biggest gains by category were in Expert Arena and Multi-Turn, as well as the Life, Physical & Social Science, Creative Writing, and Medicine professional categories.
On August 15, GLM-5.3 was announced as coming to Arena; the official preview said it would be compared with GLM-5.1 / 5.2 (the previous GLM update brought significant gains in Agent Arena).
Evidence level: Arena is a community-driven evaluation based on real-user votes and real tasks, rather than a closed laboratory benchmark; the sample and voting distribution change over time, so scores should be treated as relative reference points.
What the conclusion covers: GLM-5.2 (Max) is strongest at frontend coding (single-file HTML/React generation) and real-world agent tasks; its overall text capability is comparable to 5.1 (it is not a broad text upgrade).
Configuration: The evaluation used the Max tier; the conclusions do not apply to the low/high tiers.
Time frame: June 2026 rankings; the Frontend ranking still held in the August 4 update (before GLM-5.3 launched).
Reproduction: A same-task comparison example is available in the prompts directory under “04-X-Arena—Same-task Frontend Coding Comparison Example.”
"GLM-5.2 (Max) ranked second in Code Arena: Frontend, scoring +29 points higher than Claude Opus 4.7 (Thinking) and trai… This is a necessary excerpt; read the original source for full context.
"Agent Arena ... the strongest open-weight result we have measured, with confirmed success up +9.4% and the praise-compl… This is a necessary excerpt; read the original source for full context.
"GLM-5.2 (Max) is the strongest coding model the lab has evaluated to date."
GLM-5.2
