GLM-5.2 might be essentially the most highly effective text-only open weights LLM
seventeenth June 2026
Chinese language AI lab Z.ai launched GLM-5.2 to their coding plan subscribers on June thirteenth, after which yesterday (June sixteenth) launched the complete open weights underneath an MIT license. Comparable in measurement to their earlier GLM-5 and GLM-5.1 releases, that is 753B parameter, 1.51TB monster—with 40 energetic parameters (Combination of Specialists). GLM-5.2 is a textual content enter solely mannequin—Z.ai have a separate imaginative and prescient household most not too long ago represented by GLM-5V-Turbo, however that one isn’t open weights. GLM-5.2 has a 1 million token context window, up from GLM-5.1’s 200,000.
The thrill round this mannequin is powerful.
Synthetic Evaluation, who run some of the extensively revered unbiased benchmarks: GLM-5.2 is the brand new main open weights mannequin on the Synthetic Evaluation Intelligence Index.
GLM-5.2 is the main open weights mannequin on the Intelligence Index v4.1. At 51, it leads MiniMax-M3 (44), DeepSeek V4 Professional (max, 44) and Kimi K2.6 (43)
They did nonetheless discover it to be fairly token-hungry:
GLM-5.2 makes use of extra output tokens per job than different main open weights fashions: the mannequin makes use of 43k output tokens per Intelligence Index job, up from GLM-5.1 (26k) and above MiniMax-M3 (24k), Kimi K2.6 (35k) and DeepSeek V4 Professional (max, 37k)
The mannequin can be now ranked 2nd on the Code Area WebDev leaderboard, behind solely Claude Fable 5. That leaderboard measures “front-end net growth duties, together with agentic coding workflows”. I’m impressed to see it rank so extremely given the shortage of picture enter, which I had incorrectly assumed was a key a part of constructing a really nice frontend coding mannequin.
I’ve been making an attempt it out by way of OpenRouter, which has it from 9 completely different suppliers, virtually all of that are charging $1.40/million for enter and $4.40/million for output. For comparability, GPT-5.5 is $5/$30 and Claude Opus 4.5-4.8 is $5/$25.
Glorious pelican, disappointing opossum
GLM-5.1 gave me one in every of my favourite pelicans and my all time favourite opossum (for the immediate “Generate an SVG of a NORTH VIRGINIA OPOSSUM ON AN E-SCOOTER”.) Apparently, in each of these circumstances the mannequin selected to return SVG wrapped in an HTML doc that added further animations utilizing CSS.
Let’s attempt GLM-5.2. For “Generate an SVG of a pelican using a bicycle” I received this:
It’s a self-contained absolutely animated SVG, and the animations aren’t damaged! Typically I’ll see eyes falling off or wheels rotating independently of the bicycle however right here all the things works nice. It’s a really good vector illustration of a pelican too. Very spectacular.
Sadly, the NORTH VIRGINIA OPOSSUM ON AN E-SCOOTER didn’t come out practically as properly:
That is such a step down from GLM-5.1! As a reminder, that possum regarded like this:
5.2 didn’t even attempt to animate it.
