{"id":13004,"date":"2026-08-27T13:00:42","date_gmt":"2026-08-27T08:00:42","guid":{"rendered":"https:\/\/pivot.uz\/?p=13004"},"modified":"2026-08-27T13:03:09","modified_gmt":"2026-08-27T08:03:09","slug":"z-ai-18-mlrd-aktiv-parametrli-glm-5-3-flash-multimodal-modelini-taqdim-etdi","status":"publish","type":"post","link":"https:\/\/pivot.uz\/uz\/z-ai-18-mlrd-aktiv-parametrli-glm-5-3-flash-multimodal-modelini-taqdim-etdi\/","title":{"rendered":"Z.ai 18 mlrd aktiv parametrli GLM-5.3-Flash multimodal modelini taqdim etdi"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Xitoyning Z.ai kompaniyasi GLM <strong>turkumidagi<\/strong> yangi <strong>GLM-5.3-Flash<\/strong> sun\u2019iy intellekt modelini rasman taqdim etdi. Kompaniya uni GLM-5 seriyasidagi birinchi <strong>native multimodal<\/strong> model sifatida ta\u2019riflamoqda. Model Mixture-of-Experts (MoE) arxitekturasiga asoslangan bo\u2018lib, jami <strong>320 mlrd parametr<\/strong>ga ega, biroq har bir so\u2018rovni qayta ishlashda taxminan <strong>18 mlrd parametr faol ishlaydi<\/strong>. Bu modelni yuqori hisoblash quvvatiga ehtiyojni kamaytirgan holda murakkab vazifalarni bajarishga moslashtirishga imkon beradi.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GLM-5.3-Flash matn bilan bir qatorda tasvirlar kabi multimodal ma\u2019lumotlarni qayta ishlay oladi. Model <strong>1 million tokenlik kontekst oynasi<\/strong>ni qo\u2018llab-quvvatlaydi va dasturlash, agentlik vazifalari, hujjatlar hamda vizual ma\u2019lumotlarni tahlil qilish uchun mo\u2018ljallangan.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Z.ai ma\u2019lumotlariga ko\u2018ra, model yangi gibrid arxitekturadan foydalanadi: unda <strong>sparse attention<\/strong> va <strong>linear attention<\/strong> yondashuvlari birlashtirilgan. Kompaniya bu yechim uzoq kontekst bilan ishlashda hisoblash xarajatlarini kamaytirishini bildirgan.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GLM-5.3-Flash ommaga taqdim etilishidan oldin OpenRouter va OpenCode platformalarida <strong>Ox Alpha<\/strong> nomi ostida anonim model sifatida paydo bo\u2018lgan. Model 20-avgust kuni ishga tushirilgan va qisqa vaqt ichida dasturchilar orasida katta qiziqish uyg\u2018otgan. 26-avgust kuni Z.ai Ox Alpha aynan GLM-5.3-Flash\u2019ning oldindan sinovdan o\u2018tkazilgan versiyasi bo\u2018lganini rasman tasdiqladi. Kompaniya anonim testdan foydalanuvchilarning model haqidagi xolis fikrlarini olish uchun foydalanganini ma\u2019lum qilgan.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GLM-5.3-Flash vaznlari <strong>MIT litsenziyasi<\/strong> ostida Hugging Face platformasiga joylashtirilgan. Bu ishlab chiquvchilarga modelni yuklab olib, o\u2018z infratuzilmasida ishga tushirish va moslashtirish imkonini beradi.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Model Z.ai ekotizimida, jumladan <strong>ZCode<\/strong> muhitida ham mavjud. OpenRouter\u2019da esa GLM-5.3-Flash hozirda multimodal, kodlash va uzoq muddatli agentlik vazifalariga yo\u2018naltirilgan model sifatida ko\u2018rsatilgan.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Z.ai modelning narx va samaradorlik bo\u2018yicha oldingi GLM-5.2 bilan solishtirganda sezilarli yaxshilanganini bildirgan. Biroq bu ko\u2018rsatkichlar kompaniyaning o\u2018z benchmarklari va test metodologiyasiga asoslanganini hisobga olish kerak.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Xitoyning Z.ai kompaniyasi GLM turkumidagi yangi GLM-5.3-Flash sun\u2019iy intellekt modelini rasman taqdim etdi. Kompaniya uni GLM-5 seriyasidagi birinchi native multimodal model sifatida ta\u2019riflamoqda. Model Mixture-of-Experts (MoE) arxitekturasiga asoslangan bo\u2018lib, jami 320 mlrd parametrga ega, biroq har bir so\u2018rovni qayta ishlashda taxminan 18 mlrd parametr faol ishlaydi. Bu modelni yuqori hisoblash quvvatiga ehtiyojni kamaytirgan holda murakkab [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":13002,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"jnews-multi-image_gallery":[],"jnews_single_post":{"format":"standard"},"jnews_primary_category":[],"jnews_override_bookmark_settings":[],"jnews_paywall_metabox":[],"jnews_override_counter":[],"jnews_podcast_option":[],"jnews_podcast_series":[],"footnotes":""},"categories":[115],"tags":[],"jnews-series":[],"class_list":["post-13004","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-yangiliklar"],"_links":{"self":[{"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/posts\/13004","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/comments?post=13004"}],"version-history":[{"count":1,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/posts\/13004\/revisions"}],"predecessor-version":[{"id":13005,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/posts\/13004\/revisions\/13005"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/media\/13002"}],"wp:attachment":[{"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/media?parent=13004"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/categories?post=13004"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/tags?post=13004"},{"taxonomy":"jnews-series","embeddable":true,"href":"https:\/\/pivot.uz\/uz\/wp-json\/wp\/v2\/jnews-series?post=13004"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}