June 17, 2026
Big brain, bigger drama
GLM-5.2 is the new leading open weights model on Artificial Analysis
AI nerds are freaking out as GLM-5.2 gets smarter, cheaper, and way more chatty
TLDR: GLM-5.2 just became the top open-source AI model on a major leaderboard, giving developers a cheaper high-end option. The community is excited about the value but divided over one glaring issue: it’s powerful, pricey rivals should worry, yet it can be slow, wordy, and overloaded at launch.
A new AI model just snatched the open-source crown, and the comment section immediately turned into a mix of victory lap, price war, and server-meltdown watch party. Artificial Analysis says Z.ai’s GLM-5.2 is now the top “open weights” model — basically, a version developers can more freely run and build on — and it made a huge jump over the last release while keeping the same sticker price. It also stretched its memory window to 1 million tokens, which in normal-person language means it can keep much longer conversations and documents in mind.
But the real drama? People are split between “this changes everything” and “cool, but why is it talking so much?” One commenter called it “Opus 4.7 quality stupid prices,” saying rivals like Anthropic, OpenAI, and Google should be sweating because some providers are already selling access dirt cheap. That’s the cheerleading side.
Then came the grumbling. Several users complained GLM-5.2 seems to think forever and spend a mountain of words doing it. One person said a simple coding task took more than 15 minutes of reasoning. Another summed up the benchmark chart heartbreak with a deadpan: “That is unfortunate...” And yes, there was classic launch-day chaos: users joked the model is so popular that “their servers are melting.” The mood is basically: amazing upgrade, messy rollout, and everyone’s now arguing whether extra smarts are worth the extra yapping.
Key Points
- •Artificial Analysis says Z.ai’s GLM-5.2 is the top open-weights model on Intelligence Index v4.1 with a score of 51, ahead of MiniMax-M3, DeepSeek V4 Pro (max), and Kimi K2.6.
- •The model keeps the same parameter size as GLM-5.1 at 744B total and 40B active parameters, while improving by 11 points on the Intelligence Index.
- •The article reports strong gains across benchmarks, especially scientific reasoning metrics such as CritPt, HLE, SciCode, and GPQA Diamond, plus TerminalBench v2.1.
- •GLM-5.2 scores 1524 on GDPval-AA v2, leading open-weights models and placing roughly level with proprietary GPT-5.5 (xhigh reasoning) on that benchmark.
- •GLM-5.2 uses 43k output tokens per Intelligence Index task and costs about $0.46 per task, but the article says it still lies on the intelligence-versus-cost Pareto frontier and offers a 1M-token context window under an MIT license.