GLM-5.2 open weights: 753B text-only LLM leads open benchmarks

Z.ai released GLM-5.2 to its coding plan subscribers on June 13, and published the full open weights under an MIT license on June 16. The model is 753B parameters, occupies 1.51TB, uses 40 active parameters (Mixture of Experts), and accepts text input only with a 1 million token context window.

What is GLM-5.2 and how does it differ from earlier GLM models?

GLM-5.2 is a text-only large language model with the same approximate size family as GLM-5 and GLM-5.1 but a dramatically larger context window. It is 753B parameters, stored as 1.51TB, and configured with 40 active parameters (Mixture of Experts). Its context window is 1,000,000 tokens, up from GLM-5.1's 200,000 token window. Z.ai also maintains a separate vision family, most recently GLM-5V-Turbo, but that vision model is not available as open weights.

GLM-5.2 is distributed under an MIT license, making the full weights openly available for researchers and operators who can host the model. The release path began with a subscriber preview on June 13 and the public open-weights drop on June 16.

How does GLM-5.2 perform on benchmarks and in practice?

On the Artificial Analysis Intelligence Index v4.1, GLM-5.2 is the leading open weights model, with a score of 51, ahead of MiniMax-M3 (44), DeepSeek V4 Pro (max, 44) and Kimi K2.6 (43). Artificial Analysis also measured how many output tokens models emit per Intelligence Index task and found GLM-5.2 to be token-hungry, using 43,000 output tokens per task compared with GLM-5.1 at 26,000, MiniMax-M3 at 24,000, Kimi K2.6 at 35,000 and DeepSeek V4 Pro (max) at 37,000.

GLM-5.2 also ranked second on the Code Arena WebDev leaderboard, behind Claude Fable 5. That leaderboard is focused on front-end web development tasks and agentic coding workflows. The placement was notable given GLM-5.2 lacks image input capabilities, which might have been expected to help front-end tasks.

Operational costs reported via OpenRouter vary by provider, but most listed prices for GLM-5.2 are $1.40 per million input tokens and $4.40 per million output tokens. For context, Simon Willison noted GPT-5.5 prices at $5/$30 and Claude Opus 4.5-4.8 at $5/$25.

Simon Willison’s hands-on tests highlighted strengths and weaknesses in creative code output. The model produced a robust, fully animated SVG of a pelican riding a bicycle. It performed much worse on a separate prompt that previously produced an exemplary result in GLM-5.1: an SVG of a north Virginia opossum on an e-scooter came out poorer and without the earlier animation.

Why it matters

GLM-5.2 combines very large scale, an enormous context window, and an open MIT license. That mix lowers barriers for institutions that want to inspect, fine-tune, or self-host a top-ranking open model. Its leading position on an independent index and its high placement on a web development leaderboard suggest open weights models can now match or approach closed models on several fronts. The model’s high output token use changes operational cost math; teams that adopt it will need to plan for heavier output token billing.

What to watch

Watch how hosting and inference costs evolve for GLM-5.2 deployments and whether third-party providers change pricing as usage patterns emerge. Also watch independent benchmarks and Code Arena rankings for stability: further tests will show whether GLM-5.2’s lead and its token-hungry behavior hold across more tasks and workloads.