{"rewrite":{"id":"r_75b7b1f25c3d7d0735beb2b0","clusterId":"c_323163f03ee15d081bf27ecb","slug":"z-ai-releases-glm-5-3-as-open-model-surpassing-claude-and-gpt-in-tests","model":"deepseek-v4-flash","headline":"Z.ai Releases GLM-5.3 as Open Model, Surpassing Claude and GPT in Tests","summary":"Z.ai released its GLM-5.3 model as an open model on August 29, 2026, at 0:00 Japan time. The MoE model has 744 billion total parameters and 40 billion active parameters. Third-party organization Artificial Analysis gave it an Intelligence Index score of 60 and an Agentic Index score of 59.","whyItMatters":"GLM-5.3's open release follows Z.ai's promise from August 14 and puts a model that outperforms Claude Fable 5 and GPT-5.6 Sol in third-party tests into the open-weight ecosystem.","webCardHtml":"\u003cp\u003eZ.ai delivered on its August 14 pledge to open up GLM-5.3 once security work finished. The model went live for download on Hugging Face and ModelScope under the GLM-5.3 License.\u003c/p\u003e\u003cp\u003eThe architecture is a MoE design with 744 billion total parameters and 40 billion active parameters, which puts its compute demands at data-center scale. Unsloth has already posted a quantized GGUF version, and the UD-IQ2_M build can run on Macs with 256GB unified memory or PCs with 24GB VRAM and 256GB RAM.\u003c/p\u003e\u003cp\u003eArtificial Analysis scored the model 60 on its Intelligence Index, above Claude Opus 4.8, and 59 on its Agentic Index, above GPT-5.6 Sol and Grok 4.6. The smaller GLM-5.3-Flash was released as an open model on August 26.\u003c/p\u003e","blueskyPost":"Z.ai's GLM-5.3 launches with 744B total parameters but only 40B active, a design that keeps inference costs low. The open release puts pressure on closed rivals to justify their pricing.","twitterPost":"GLM-5.3's 40B active parameters make it cheap to run, but its open license is the real threat to closed models.","threadsPost":"Z.ai's GLM-5.3 is open and free, but the engineering choice matters: 744B parameters with only 40B active keeps costs down. The open release challenges Claude and GPT's paid tiers, forcing a pricing conversation.","newsletterBlurb":"Z.ai made good on its promise and released GLM-5.3 as an open model at midnight Japan time on August 29. The 744-billion-parameter MoE model scored 60 on Artificial Analysis' Intelligence Index and 59 on its Agentic Index, topping Claude Opus 4.8, GPT-5.6 Sol, and Grok 4.6 in those measures.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260829-glm-5-3-open/\",\"title\":\"Chinese AI Model 'GLM-5.3' Released as Open Model as Promised, High-Performance Model Rivaling Claude Fable 5 and GPT-5.6 Sol\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4861,"outputTokens":732,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1788023874,"createdAt":"2026-08-29T17:15:16.000Z","publishedAt":"2026-08-29T17:16:44.000Z","updatedAt":"2026-08-29T17:15:16.000Z"},"cluster":{"id":"c_323163f03ee15d081bf27ecb","canonicalTitle":"中国製AIモデル「GLM-5.3」が約束通りオープンモデルとして無償公開される、Claude Fable 5やGPT-5.6 Solに匹敵する高性能モデル","representativeArticleId":"a_5a63c8030fed79f30c45cd6d","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"GLM-5.3\"],\"studios\":[\"Z.ai\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-08-28T15:10:00.000Z","lastSeenAt":"2026-08-28T15:10:00.000Z","updatedAt":"2026-08-29T17:16:45.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260829-glm-5-3-open/","title":"中国製AIモデル「GLM-5.3」が約束通りオープンモデルとして無償公開される、Claude Fable 5やGPT-5.6 Solに匹敵する高性能モデル"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["GLM-5.3"],"studios":["Z.ai"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}