[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:sr":3,"public-menus:all":38,"post:qwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning:sr":205,"related:post:qwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning:sr:1":700},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","sr","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,181,193],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-folder","custom",[],{"id":176,"title":177,"url":179,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":180},"item-28",{"de":178,"en":178,"es":178,"fr":178,"it":178,"ru":178,"sr":178,"zh":178},"Solr Suggester","\u002Fportfolio\u002Fsolr-fuzzy-suggester-und-solr-infix-suggester-abfrage-ueber-ajax-und-filterung",[],{"id":182,"title":183,"url":191,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":192},"item-27",{"de":184,"en":185,"es":186,"fr":187,"it":188,"ru":189,"sr":190,"zh":185},"Firmenwebseite SEO","Company Website SEO","Sitio web corporativo SEO","Site web d’entreprise SEO","Sito web aziendale SEO","Корпоративный сайт SEO","Пословна веб-страница SEO","\u002Fportfolio\u002Fseo-sem-branding-mobile-webseite-muenchen",[],{"id":194,"title":195,"url":203,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":204},"item-31",{"de":196,"en":197,"es":198,"fr":199,"it":200,"ru":201,"sr":202,"zh":197},"Digitalisierungsportal","Digitalization Portal","Portal de digitalización","Portail de numérisation","Portale di digitalizzazione","Портал цифровизации","Портал за дигитализацију","\u002Fportfolio\u002Fdigitalisierungsportal-archiv-museum-bibliothek-ead-lido-mets-mods",[],{"statusCode":4,"data":206,"message":699},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":387,"featuredImage":388,"featuredImageAlt":389,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":390,"publishedAt":391,"createdAt":392,"updatedAt":393,"seoLocalePaths":394,"categories":403,"author":424,"translations":429},"445","Qwen 3.6 u produkciji: Runbook za izdavanje, AI rollback i LLMOps verziranje","qwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u003Cp>\u003Cb>Qwen3.6-Plus\u003C\u002Fb> je važan jer pomera Qwen liniju od obećavajućeg agentskog modela ka nečemu što je mnogo bliže izvršnom sloju produkcionog nivoa. Raniji članak o Qwen 3.5-Plus na stajic.de je već ispravno uokvirio tu promenu: tržište se udaljava od inteligencije zasnovane samo na ćaskanju i kreće se ka pouzdanom izvršavanju u više koraka. Qwen3.6 ide dalje u tom pravcu sa snažnijim agentskim kodiranjem, boljim multimodalnim zaključivanjem i stavom o objavljivanju koji je više fokusiran na stabilnost.\u003C\u002Fp>\n\u003Cp>To čini ovu temu prirodnim izborom za \u003Cb>Enterprise Delivery OS\u003C\u002Fb>. Ona prvenstveno pripada \u003Cb>LLMOps Playbook-u\u003C\u002Fb>, sa najjačim podudaranjem pod \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. Istovremeno, ona bi takođe trebalo prirodno da stoji unutar \u003Cb>Release Runbook\u003C\u002Fb> i \u003Cb>AI Rollback Runbook\u003C\u002Fb>, jer nadogradnja modela nije samo izbor modela. To je događaj objavljivanja, scenario povraćaja (rollback) i problem verziranja u isto vreme.\u003C\u002Fp>\n\u003Cp>Zvanično lansiranje Qwen3.6-Plus pozicionira model kao značajnu nadogradnju u odnosu na Qwen3.5-Plus, posebno za agentsko kodiranje, rešavanje problema na nivou repozitorijuma, multimodalno zaključivanje i stabilno izvršavanje u stvarnom svetu. Alibaba takođe navodi da je hostovani Plus model dostupan odmah sa podrazumevanim kontekstnim prozorom od 1M, dok varijante Qwen3.6 sa otvorenim težinama (open-weight) proširuju porodicu za timove koji žele veću kontrolu nad izborom implementacije i inferencije.\u003C\u002Fp>\n\u003Ch2>Šta se promenilo od Qwen 3.5 do Qwen 3.6\u003C\u002Fh2>\n\u003Cp>Originalni članak o Qwen 3.5-Plus na stajic.de fokusirao se na četiri praktične prednosti: veliki kontekst, ponašanje pri korišćenju alata, multimodalne sposobnosti i kretanje ka pouzdanom agentskom izvršavanju. Qwen3.6-Plus zadržava tu osnovu, ali zvanično izdanje izoštrava operativnu vrednost. Stavlja mnogo veći naglasak na kvalitet agentskog kodiranja, izvršavanje u terminalskom stilu, dugoročno korišćenje alata i jaču stabilnost zasnovanu na povratnim informacijama o implementaciji iz Qwen3.5 ere.\u003C\u002Fp>\n\u003Cul>\u003Cli>Kontekstni prozor od 1M podrazumevano u hostovanom Plus modelu\u003C\u002Fli>\u003Cli>Značajno poboljšana sposobnost agentskog kodiranja\u003C\u002Fli>\u003Cli>Bolja multimodalna percepcija i zaključivanje\u003C\u002Fli>\u003Cli>Stabilnija i pouzdanija osnova za stvarne radne tokove programera\u003C\u002Fli>\u003Cli>Šira Qwen3.6 porodica koja takođe uključuje varijante sa otvorenim težinama za slučajeve samostalnog hostovanja\u003C\u002Fli>\u003C\u002Ful>\n\u003Cpre class=\"code-block\">\u003Ccode>\u002F\u002F Minimalna ideja za migraciju\nconst modelConfig = { provider: "qwen", model: "qwen3.6-plus", maxContext: 1000000, mode: "agentic-coding", tools: ["browser", "bash", "search", "file-edit"]\n}; \u002F\u002F Težak deo nije promena modela.\n\u002F\u002F Težak deo su kontrola objavljivanja, evaluacija i spremnost za povraćaj (rollback).\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Zašto je Qwen 3.6 tema za Release Runbook\u003C\u002Fh2>\n\u003Cp>Živi \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\">Release Runbook\u003C\u002Fa> definiše bezbednost objavljivanja kroz provere pre leta (preflight checks), jasne vlasnike, verifikaciju u odnosu na kriterijume prihvatanja, prikupljene dokaze i pregled nakon objavljivanja. Produkciona nadogradnja sa Qwen3.5-Plus na Qwen3.6-Plus se tačno uklapa u taj obrazac. Objavljivanje modela nije samo nova funkcija. To je promena ponašanja unutar sistema uživo, a to znači da zaslužuje disciplinu nivoa objavljivanja (release-grade).\u003C\u002Fp>\n\u003Cp>Ovo postaje još važnije kada se model koristi za generisanje koda, izvršavanje alata, zaključivanje na nivou repozitorijuma ili multimodalne radne tokove. Što je veća operativna površina, to postaje opasnije tretirati nadogradnju kao jedno podešavanje konfiguracije.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Kontrolna lista za objavljivanje nadogradnje modela\n1. Definišite ciljnu verziju i obim implementacije\n2. Zamrznite promene promptova i rutiranja tokom validacije\n3. Pokrenite sistem za evaluaciju na stabilnom testnom skupu\n4. Verifikujte delte troškova, latencije i stope neuspeha\n5. Potvrdite putanju povraćaja (rollback) i pragove za aktiviranje povraćaja\n6. Odobrite objavljivanje sa imenovanim vlasnikom i paketom dokaza\n7. Pratite ponašanje nakon objavljivanja pre potpunog prebacivanja saobraćaja\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Zašto je Qwen 3.6 takođe tema za AI Rollback\u003C\u002Fh2>\n\u003Cp>Živi \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Fai-rollback\">AI Rollback Runbook\u003C\u002Fa> je eksplicitan: LLM sistemi mogu nazadovati kroz promene promptova, promene rutiranja, ažuriranja modela ili pomeranje podataka (data drift). To nije teorijski. Nadogradnja modela može poboljšati benchmark-ove kodiranja, a ipak dovesti do regresije produkcionog radnog toka koji zavisi od stabilnosti formatiranja, discipline alata, bezbednosnog ponašanja, profila troškova ili stila izlaza.\u003C\u002Fp>\n\u003Cp>Qwen3.6-Plus može biti jači u celini, ali produkcioni sistemi ne otkazuju na prosečnom kvalitetu. Oni otkazuju na graničnim ponašanjima, skrivenim zavisnostima i krhkim pretpostavkama integracije. Zato je svakoj nadogradnji modela potrebna eksplicitna uslovljenost povraćaja (rollback) pre nego što se saobraćaj preusmeri.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>{ "rollbackTriggers": { "qualityDropPct": 5, "toolCallFailurePct": 2, "costIncreasePct": 20, "latencyIncreasePct": 25, "safetyViolationCount": 1 }, "rollbackTarget": "qwen3.5-plus", "freezeWindow": "24h", "requiredEvidence": [ "eval-report", "traffic-split-report", "post-release-verification" ]\n}\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Verziranje promptova i modela je kontrolna površina\u003C\u002Fh2>\n\u003Cp>Živa stranica \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\">Versioning (Prompts, Models)\u003C\u002Fa> daje pravi okvir: promptovima i modelima je potrebna sledljivost i kontrolisana promena. Ta tačka postaje mnogo konkretnija kada se porodica modela brzo nadograđuje. Ako tim ne može da kaže koji set promptova, logika rutiranja, politika temperature, dozvole alata i osnovna linija evaluacije su bili aktivni pri određenom objavljivanju, onda sistem nije zaista verziran. On je samo konfigurisan.\u003C\u002Fp>\n\u003Cp>Zapis verzije produkcionog nivoa treba da poveže identitet modela, paket promptova, politiku alata, skup za evaluaciju i odluku o objavljivanju. Ovo je posebno važno za Qwen3.6 jer je model dizajniran za snažnije agentsko izvršavanje. Sposobnije ponašanje znači veću potrebu za eksplicitnim granicama verzija.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>{ "versionId": "llm-stack-2026-05-04-a", "model": "qwen3.6-plus", "fallbackModel": "qwen3.5-plus", "promptBundle": "repo-agent-v12", "toolPolicy": "repo-agent-safe-tools-v4", "routerPolicy": "coding-heavy-workloads-v3", "evalSet": "agentic-coding-regression-suite-v7", "approvedBy": "llmops-owner", "releaseState": "canary"\n}\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Zašto je najbolji primarni smeštaj LLMOps priručnik\u003C\u002Fh2>\n\u003Cp>Aktivni \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\">LLMOps priručnik\u003C\u002Fa> već definiše ključnu operativnu logiku: verziranje promptova i modela, evaluaciju sa kapijama kvaliteta, uvođenje kroz canary ili A\u002FB putanje, praćenje odstupanja ili regresija i održavanje brzog povratka na prethodnu verziju (rollback). Qwen3.6-Plus je skoro školski primer zašto taj priručnik postoji. Model je jači, ali nadogradnja postaje vredna samo kada ponašanje ostane stabilno kroz promene.\u003C\u002Fp>\n\u003Cul>\u003Cli>Verziranje štiti sledljivost i kontrolisanu promenu\u003C\u002Fli>\u003Cli>Evaluacioni okvir štiti kvalitet pre preusmeravanja saobraćaja\u003C\u002Fli>\u003Cli>Canary i A\u002FB izdanja smanjuju opseg uticaja nadogradnje modela\u003C\u002Fli>\u003Cli>Monitoring hvata regresije koje je statička evaluacija propustila\u003C\u002Fli>\u003Cli>Strategija povratka (rollback) održava sistem reverzibilnim kada stvarni saobraćaj otkrije slabosti\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>Ovo je takođe razlog zašto se aktivna stranica \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fcanary-and-ab\">Canary i A\u002FB izdanja\u003C\u002Fa> prirodno uklapa ovde. Sposoban model ne bi trebalo da ide direktno od uzbuđenja zbog benchmark rezultata do punog produkcionog saobraćaja. Sigurniji obrazac je fazno uvođenje sa eksplicitnim dokazima u svakoj fazi.\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>Redosled LLMOps uvođenja\n1. Fiksirajte tačan model i paket promptova\n2. Pokrenite oflajn evaluaciju na fiksnom setu za regresiju\n3. Započnite canary saobraćaj sa eksplicitnim zaštitnim ogradama\n4. Uporedite metriku kvaliteta, latencije, cene i neuspeha alata\n5. Proširite saobraćaj samo ako se pragovi održavaju\n6. Održavajte povratak (rollback) trenutnim i dokumentovanim\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>Gde Qwen 3.6 stvara pravu prednost\u003C\u002Fh2>\n\u003Cp>Najjača praktična vrednost Qwen3.6 nije u tome što zvuči pametnije u demou. Prava prednost dolazi tamo gde je kontinuitet radnog toka važan: kodiranje na nivou repozitorijuma, dugoročno korišćenje alata, multimodalno debagovanje i ponovljeno izvršavanje pod promenljivim kontekstom. To je mesto gde agentniji model može ukloniti stvarno trenje umesto da samo poboljša glavne benchmark rezultate.\u003C\u002Fp>\n\u003Cul>\u003Cli>Zadaci kodiranja na nivou repozitorijuma sa više datoteka i dužim lancima zavisnosti\u003C\u002Fli>\u003Cli>Putanje izvršavanja orijentisane na terminal gde je disciplina alata važna\u003C\u002Fli>\u003Cli>Multimodalni QA i UI debagovanje gde su snimci ekrana i dokumenti važni\u003C\u002Fli>\u003Cli>Ops i analiza incidenata sa većim kontekstom i izvršavanjem u stilu priručnika\u003C\u002Fli>\u003Cli>Agentni radni tokovi gde je stabilnost kroz mnogo koraka važnija od briljantnosti u jednom odgovoru\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>Taj pravac je u skladu i sa lansiranjem Qwen3.6-Plus i sa open-weight Qwen3.6 izdanjima, koja naglašavaju agentno kodiranje, rezonovanje o repozitorijumu, očuvanje razmišljanja i širu fleksibilnost uvođenja. Za timove koji već testiraju Qwen3.5-Plus, pitanje više nije da li je Qwen3.6 zanimljiv. Pravo pitanje je da li tim može da ga nadogradi sa istom disciplinom koja se koristi za bilo koju drugu produkcionu zavisnost.\u003C\u002Fp>\n\u003Ch2>Kompromisi koje treba poštovati pre nadogradnje\u003C\u002Fh2>\n\u003Cul>\u003Cli>Veći kontekstni prozor ne uklanja potrebu za strukturiranim unosima i planiranjem pretrage\u003C\u002Fli>\u003Cli>Jače agentno kodiranje povećava potrebu za politikom alata, sandboxing-om i logovima koji se mogu ponovo reprodukovati\u003C\u002Fli>\u003Cli>Hostovane i open-weight varijante stvaraju različite kompromise u pogledu izdanja, privatnosti i operacija\u003C\u002Fli>\u003Cli>Bolji prosečan model i dalje može unazaditi jedan kritičan radni tok ako je evaluacija slaba\u003C\u002Fli>\u003Cli>Odstupanje verzija kroz prompte, rutere i krajnje tačke modela može uništiti sledljivost ako se ostavi bez kontrole\u003C\u002Fli>\u003C\u002Ful>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">Nadogradnja modela je nadogradnja sposobnosti samo ako operativni sistem oko njega može da dokaže stabilnost, prati promenu i brzo je poništi.\u003Ccite class=\"block mt-2 text-sm\">— LLMOps perspektiva\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Ch2>Najbolji smeštaj u stubove\u003C\u002Fh2>\n\u003Cp>Najbolji primarni smeštaj za ovaj članak je \u003Cb>LLMOps priručnik\u003C\u002Fb>, sa najjačim pod-uklapanjem pod \u003Cb>Verziranje (promptovi, modeli)\u003C\u002Fb>. Dodatni smeštaj pod \u003Cb>Priručnik za izdanja\u003C\u002Fb> i \u003Cb>Priručnik za AI povratak\u003C\u002Fb> je opravdan jer je tema eksplicitno o kontrolisanom izdanju, spremnosti za povratak i sledljivosti verzija modela. Drugim rečima, Qwen3.6 nije samo priča o modelu. To je operativna priča.\u003C\u002Fp>\n\u003Ch2>Završna perspektiva\u003C\u002Fh2>\n\u003Cp>Qwen3.6-Plus je snažan signal da agentna veštačka inteligencija postaje praktičnija za ozbiljan inženjerski rad. Ali pravi signal zrelosti nije grafikon benchmarka. To je da li tim može da izda novi model kroz priručnik zasnovan na dokazima, očuva sledljivost verzija prompta i modela, bezbedno sprovede canary promenu, prati stvarno ponašanje i brzo se vrati na prethodno stanje ako jedan kritičan radni tok nazaduje. To je razlika između eksperimentisanja sa modelima i upravljanja njima.\u003C\u002Fp>\n\u003Cdiv class=\"ce-delimiter cdx-block my-8\">\u003C\u002Fdiv>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fblog\u002Fnew-qwen-3-5-plus\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">Novi Qwen 3.5-Plus: Open-source AI sada postaje ozbiljan\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Raniji članak o Qwen 3.5-Plus kao pomaku sa inteligencije u stilu ćaskanja ka pouzdanijem agentskom izvršavanju.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">LLMOps priručnik\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Održavajte ponašanje LLM-a stabilnim kroz promene putem verziranja, evaluacije, kanarinskih izdanja, monitoringa i procedura brzog vraćanja unazad.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">Verziranje (promptovi, modeli)\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Strategija verziranja za promptove i modele kako bi se osigurala sledljivost i kontrolisana promena.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">Runbook za izdanja\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Koristite provere pre leta, imenovane vlasnike, verifikaciju prema kriterijumima prihvatanja, zabeležene dokaze i pregled nakon izdanja.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Fai-rollback\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">Runbook za AI rollback\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">LLM sistemi mogu nazadovati kroz promene promptova, promene rutiranja, ažuriranja modela ili pomeranje podataka. Zamrznite, verifikujte, vratite unazad i učite.\u003C\u002Fp>\u003C\u002Fa>",{"time":212,"blocks":213,"version":386},1777884888545,[214,218,221,224,228,231,241,245,248,251,254,257,260,263,266,269,272,275,278,281,284,287,295,298,301,304,307,315,318,321,329,334,337,340,343,346,349,358,365,372,379],{"data":215,"type":217},{"text":216},"\u003Cb>Qwen3.6-Plus\u003C\u002Fb> je važan jer pomera Qwen liniju od obećavajućeg agentskog modela ka nečemu što je mnogo bliže izvršnom sloju produkcionog nivoa. Raniji članak o Qwen 3.5-Plus na stajic.de je već ispravno uokvirio tu promenu: tržište se udaljava od inteligencije zasnovane samo na ćaskanju i kreće se ka pouzdanom izvršavanju u više koraka. Qwen3.6 ide dalje u tom pravcu sa snažnijim agentskim kodiranjem, boljim multimodalnim zaključivanjem i stavom o objavljivanju koji je više fokusiran na stabilnost.","paragraph",{"data":219,"type":217},{"text":220},"To čini ovu temu prirodnim izborom za \u003Cb>Enterprise Delivery OS\u003C\u002Fb>. Ona prvenstveno pripada \u003Cb>LLMOps Playbook-u\u003C\u002Fb>, sa najjačim podudaranjem pod \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. Istovremeno, ona bi takođe trebalo prirodno da stoji unutar \u003Cb>Release Runbook\u003C\u002Fb> i \u003Cb>AI Rollback Runbook\u003C\u002Fb>, jer nadogradnja modela nije samo izbor modela. To je događaj objavljivanja, scenario povraćaja (rollback) i problem verziranja u isto vreme.",{"data":222,"type":217},{"text":223},"Zvanično lansiranje Qwen3.6-Plus pozicionira model kao značajnu nadogradnju u odnosu na Qwen3.5-Plus, posebno za agentsko kodiranje, rešavanje problema na nivou repozitorijuma, multimodalno zaključivanje i stabilno izvršavanje u stvarnom svetu. Alibaba takođe navodi da je hostovani Plus model dostupan odmah sa podrazumevanim kontekstnim prozorom od 1M, dok varijante Qwen3.6 sa otvorenim težinama (open-weight) proširuju porodicu za timove koji žele veću kontrolu nad izborom implementacije i inferencije.",{"data":225,"type":42},{"text":226,"level":227},"Šta se promenilo od Qwen 3.5 do Qwen 3.6",2,{"data":229,"type":217},{"text":230},"Originalni članak o Qwen 3.5-Plus na stajic.de fokusirao se na četiri praktične prednosti: veliki kontekst, ponašanje pri korišćenju alata, multimodalne sposobnosti i kretanje ka pouzdanom agentskom izvršavanju. Qwen3.6-Plus zadržava tu osnovu, ali zvanično izdanje izoštrava operativnu vrednost. Stavlja mnogo veći naglasak na kvalitet agentskog kodiranja, izvršavanje u terminalskom stilu, dugoročno korišćenje alata i jaču stabilnost zasnovanu na povratnim informacijama o implementaciji iz Qwen3.5 ere.",{"data":232,"type":240},{"items":233,"style":239},[234,235,236,237,238],"Kontekstni prozor od 1M podrazumevano u hostovanom Plus modelu","Značajno poboljšana sposobnost agentskog kodiranja","Bolja multimodalna percepcija i zaključivanje","Stabilnija i pouzdanija osnova za stvarne radne tokove programera","Šira Qwen3.6 porodica koja takođe uključuje varijante sa otvorenim težinama za slučajeve samostalnog hostovanja","unordered","list",{"data":242,"type":244},{"code":243},"\u002F\u002F Minimalna ideja za migraciju\nconst modelConfig = { provider: \"qwen\", model: \"qwen3.6-plus\", maxContext: 1000000, mode: \"agentic-coding\", tools: [\"browser\", \"bash\", \"search\", \"file-edit\"]\n}; \u002F\u002F Težak deo nije promena modela.\n\u002F\u002F Težak deo su kontrola objavljivanja, evaluacija i spremnost za povraćaj (rollback).","code",{"data":246,"type":42},{"text":247,"level":227},"Zašto je Qwen 3.6 tema za Release Runbook",{"data":249,"type":217},{"text":250},"Živi \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\">Release Runbook\u003C\u002Fa> definiše bezbednost objavljivanja kroz provere pre leta (preflight checks), jasne vlasnike, verifikaciju u odnosu na kriterijume prihvatanja, prikupljene dokaze i pregled nakon objavljivanja. Produkciona nadogradnja sa Qwen3.5-Plus na Qwen3.6-Plus se tačno uklapa u taj obrazac. Objavljivanje modela nije samo nova funkcija. To je promena ponašanja unutar sistema uživo, a to znači da zaslužuje disciplinu nivoa objavljivanja (release-grade).",{"data":252,"type":217},{"text":253},"Ovo postaje još važnije kada se model koristi za generisanje koda, izvršavanje alata, zaključivanje na nivou repozitorijuma ili multimodalne radne tokove. Što je veća operativna površina, to postaje opasnije tretirati nadogradnju kao jedno podešavanje konfiguracije.",{"data":255,"type":244},{"code":256},"Kontrolna lista za objavljivanje nadogradnje modela\n1. Definišite ciljnu verziju i obim implementacije\n2. Zamrznite promene promptova i rutiranja tokom validacije\n3. Pokrenite sistem za evaluaciju na stabilnom testnom skupu\n4. Verifikujte delte troškova, latencije i stope neuspeha\n5. Potvrdite putanju povraćaja (rollback) i pragove za aktiviranje povraćaja\n6. Odobrite objavljivanje sa imenovanim vlasnikom i paketom dokaza\n7. Pratite ponašanje nakon objavljivanja pre potpunog prebacivanja saobraćaja",{"data":258,"type":42},{"text":259,"level":227},"Zašto je Qwen 3.6 takođe tema za AI Rollback",{"data":261,"type":217},{"text":262},"Živi \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Fai-rollback\">AI Rollback Runbook\u003C\u002Fa> je eksplicitan: LLM sistemi mogu nazadovati kroz promene promptova, promene rutiranja, ažuriranja modela ili pomeranje podataka (data drift). To nije teorijski. Nadogradnja modela može poboljšati benchmark-ove kodiranja, a ipak dovesti do regresije produkcionog radnog toka koji zavisi od stabilnosti formatiranja, discipline alata, bezbednosnog ponašanja, profila troškova ili stila izlaza.",{"data":264,"type":217},{"text":265},"Qwen3.6-Plus može biti jači u celini, ali produkcioni sistemi ne otkazuju na prosečnom kvalitetu. Oni otkazuju na graničnim ponašanjima, skrivenim zavisnostima i krhkim pretpostavkama integracije. Zato je svakoj nadogradnji modela potrebna eksplicitna uslovljenost povraćaja (rollback) pre nego što se saobraćaj preusmeri.",{"data":267,"type":244},{"code":268},"{ \"rollbackTriggers\": { \"qualityDropPct\": 5, \"toolCallFailurePct\": 2, \"costIncreasePct\": 20, \"latencyIncreasePct\": 25, \"safetyViolationCount\": 1 }, \"rollbackTarget\": \"qwen3.5-plus\", \"freezeWindow\": \"24h\", \"requiredEvidence\": [ \"eval-report\", \"traffic-split-report\", \"post-release-verification\" ]\n}",{"data":270,"type":42},{"text":271,"level":227},"Verziranje promptova i modela je kontrolna površina",{"data":273,"type":217},{"text":274},"Živa stranica \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\">Versioning (Prompts, Models)\u003C\u002Fa> daje pravi okvir: promptovima i modelima je potrebna sledljivost i kontrolisana promena. Ta tačka postaje mnogo konkretnija kada se porodica modela brzo nadograđuje. Ako tim ne može da kaže koji set promptova, logika rutiranja, politika temperature, dozvole alata i osnovna linija evaluacije su bili aktivni pri određenom objavljivanju, onda sistem nije zaista verziran. On je samo konfigurisan.",{"data":276,"type":217},{"text":277},"Zapis verzije produkcionog nivoa treba da poveže identitet modela, paket promptova, politiku alata, skup za evaluaciju i odluku o objavljivanju. Ovo je posebno važno za Qwen3.6 jer je model dizajniran za snažnije agentsko izvršavanje. Sposobnije ponašanje znači veću potrebu za eksplicitnim granicama verzija.",{"data":279,"type":244},{"code":280},"{ \"versionId\": \"llm-stack-2026-05-04-a\", \"model\": \"qwen3.6-plus\", \"fallbackModel\": \"qwen3.5-plus\", \"promptBundle\": \"repo-agent-v12\", \"toolPolicy\": \"repo-agent-safe-tools-v4\", \"routerPolicy\": \"coding-heavy-workloads-v3\", \"evalSet\": \"agentic-coding-regression-suite-v7\", \"approvedBy\": \"llmops-owner\", \"releaseState\": \"canary\"\n}",{"data":282,"type":42},{"text":283,"level":227},"Zašto je najbolji primarni smeštaj LLMOps priručnik",{"data":285,"type":217},{"text":286},"Aktivni \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\">LLMOps priručnik\u003C\u002Fa> već definiše ključnu operativnu logiku: verziranje promptova i modela, evaluaciju sa kapijama kvaliteta, uvođenje kroz canary ili A\u002FB putanje, praćenje odstupanja ili regresija i održavanje brzog povratka na prethodnu verziju (rollback). Qwen3.6-Plus je skoro školski primer zašto taj priručnik postoji. Model je jači, ali nadogradnja postaje vredna samo kada ponašanje ostane stabilno kroz promene.",{"data":288,"type":240},{"items":289,"style":239},[290,291,292,293,294],"Verziranje štiti sledljivost i kontrolisanu promenu","Evaluacioni okvir štiti kvalitet pre preusmeravanja saobraćaja","Canary i A\u002FB izdanja smanjuju opseg uticaja nadogradnje modela","Monitoring hvata regresije koje je statička evaluacija propustila","Strategija povratka (rollback) održava sistem reverzibilnim kada stvarni saobraćaj otkrije slabosti",{"data":296,"type":217},{"text":297},"Ovo je takođe razlog zašto se aktivna stranica \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fcanary-and-ab\">Canary i A\u002FB izdanja\u003C\u002Fa> prirodno uklapa ovde. Sposoban model ne bi trebalo da ide direktno od uzbuđenja zbog benchmark rezultata do punog produkcionog saobraćaja. Sigurniji obrazac je fazno uvođenje sa eksplicitnim dokazima u svakoj fazi.",{"data":299,"type":244},{"code":300},"Redosled LLMOps uvođenja\n1. Fiksirajte tačan model i paket promptova\n2. Pokrenite oflajn evaluaciju na fiksnom setu za regresiju\n3. Započnite canary saobraćaj sa eksplicitnim zaštitnim ogradama\n4. Uporedite metriku kvaliteta, latencije, cene i neuspeha alata\n5. Proširite saobraćaj samo ako se pragovi održavaju\n6. Održavajte povratak (rollback) trenutnim i dokumentovanim",{"data":302,"type":42},{"text":303,"level":227},"Gde Qwen 3.6 stvara pravu prednost",{"data":305,"type":217},{"text":306},"Najjača praktična vrednost Qwen3.6 nije u tome što zvuči pametnije u demou. Prava prednost dolazi tamo gde je kontinuitet radnog toka važan: kodiranje na nivou repozitorijuma, dugoročno korišćenje alata, multimodalno debagovanje i ponovljeno izvršavanje pod promenljivim kontekstom. To je mesto gde agentniji model može ukloniti stvarno trenje umesto da samo poboljša glavne benchmark rezultate.",{"data":308,"type":240},{"items":309,"style":239},[310,311,312,313,314],"Zadaci kodiranja na nivou repozitorijuma sa više datoteka i dužim lancima zavisnosti","Putanje izvršavanja orijentisane na terminal gde je disciplina alata važna","Multimodalni QA i UI debagovanje gde su snimci ekrana i dokumenti važni","Ops i analiza incidenata sa većim kontekstom i izvršavanjem u stilu priručnika","Agentni radni tokovi gde je stabilnost kroz mnogo koraka važnija od briljantnosti u jednom odgovoru",{"data":316,"type":217},{"text":317},"Taj pravac je u skladu i sa lansiranjem Qwen3.6-Plus i sa open-weight Qwen3.6 izdanjima, koja naglašavaju agentno kodiranje, rezonovanje o repozitorijumu, očuvanje razmišljanja i širu fleksibilnost uvođenja. Za timove koji već testiraju Qwen3.5-Plus, pitanje više nije da li je Qwen3.6 zanimljiv. Pravo pitanje je da li tim može da ga nadogradi sa istom disciplinom koja se koristi za bilo koju drugu produkcionu zavisnost.",{"data":319,"type":42},{"text":320,"level":227},"Kompromisi koje treba poštovati pre nadogradnje",{"data":322,"type":240},{"items":323,"style":239},[324,325,326,327,328],"Veći kontekstni prozor ne uklanja potrebu za strukturiranim unosima i planiranjem pretrage","Jače agentno kodiranje povećava potrebu za politikom alata, sandboxing-om i logovima koji se mogu ponovo reprodukovati","Hostovane i open-weight varijante stvaraju različite kompromise u pogledu izdanja, privatnosti i operacija","Bolji prosečan model i dalje može unazaditi jedan kritičan radni tok ako je evaluacija slaba","Odstupanje verzija kroz prompte, rutere i krajnje tačke modela može uništiti sledljivost ako se ostavi bez kontrole",{"data":330,"type":333},{"text":331,"caption":332},"Nadogradnja modela je nadogradnja sposobnosti samo ako operativni sistem oko njega može da dokaže stabilnost, prati promenu i brzo je poništi.","LLMOps perspektiva","quote",{"data":335,"type":42},{"text":336,"level":227},"Najbolji smeštaj u stubove",{"data":338,"type":217},{"text":339},"Najbolji primarni smeštaj za ovaj članak je \u003Cb>LLMOps priručnik\u003C\u002Fb>, sa najjačim pod-uklapanjem pod \u003Cb>Verziranje (promptovi, modeli)\u003C\u002Fb>. Dodatni smeštaj pod \u003Cb>Priručnik za izdanja\u003C\u002Fb> i \u003Cb>Priručnik za AI povratak\u003C\u002Fb> je opravdan jer je tema eksplicitno o kontrolisanom izdanju, spremnosti za povratak i sledljivosti verzija modela. Drugim rečima, Qwen3.6 nije samo priča o modelu. To je operativna priča.",{"data":341,"type":42},{"text":342,"level":227},"Završna perspektiva",{"data":344,"type":217},{"text":345},"Qwen3.6-Plus je snažan signal da agentna veštačka inteligencija postaje praktičnija za ozbiljan inženjerski rad. Ali pravi signal zrelosti nije grafikon benchmarka. To je da li tim može da izda novi model kroz priručnik zasnovan na dokazima, očuva sledljivost verzija prompta i modela, bezbedno sprovede canary promenu, prati stvarno ponašanje i brzo se vrati na prethodno stanje ako jedan kritičan radni tok nazaduje. To je razlika između eksperimentisanja sa modelima i upravljanja njima.",{"data":347,"type":348},{},"delimiter",{"data":350,"type":357},{"link":351,"meta":352},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fblog\u002Fnew-qwen-3-5-plus",{"image":353,"title":355,"description":356},{"url":354},"","Novi Qwen 3.5-Plus: Open-source AI sada postaje ozbiljan","Raniji članak o Qwen 3.5-Plus kao pomaku sa inteligencije u stilu ćaskanja ka pouzdanijem agentskom izvršavanju.","linkTool",{"data":359,"type":357},{"link":360,"meta":361},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops",{"image":362,"title":363,"description":364},{"url":354},"LLMOps priručnik","Održavajte ponašanje LLM-a stabilnim kroz promene putem verziranja, evaluacije, kanarinskih izdanja, monitoringa i procedura brzog vraćanja unazad.",{"data":366,"type":357},{"link":367,"meta":368},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning",{"image":369,"title":370,"description":371},{"url":354},"Verziranje (promptovi, modeli)","Strategija verziranja za promptove i modele kako bi se osigurala sledljivost i kontrolisana promena.",{"data":373,"type":357},{"link":374,"meta":375},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Frelease-runbook",{"image":376,"title":377,"description":378},{"url":354},"Runbook za izdanja","Koristite provere pre leta, imenovane vlasnike, verifikaciju prema kriterijumima prihvatanja, zabeležene dokaze i pregled nakon izdanja.",{"data":380,"type":357},{"link":381,"meta":382},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fenterprise\u002Frunbooks\u002Fai-rollback",{"image":383,"title":384,"description":385},{"url":354},"Runbook za AI rollback","LLM sistemi mogu nazadovati kroz promene promptova, promene rutiranja, ažuriranja modela ili pomeranje podataka. Zamrznite, verifikujte, vratite unazad i učite.","2.31","Qwen 3.6 nije samo još jedna nadogradnja modela. To je istovremeno događaj objavljivanja, scenario povratka na prethodnu verziju i problem verziranja. Ovaj članak objašnjava kako Qwen 3.6 treba tretirati u produkciji kroz LLMOps disciplinu, sledljivost promptova i modela, kontrolisano uvođenje i spremnost za povratak na prethodnu verziju zasnovanu na dokazima.","\u002Fuploads\u002F2026\u002F02\u002Fnew-qwen-3-5-plus-1771515512741-dcbi9p.webp","new-qwen-3-5-plus-1771515512741-dcbi9p","PUBLISHED","2026-05-04T02:49:00.000Z","2026-05-04T08:49:10.349Z","2026-05-04T09:43:23.744Z",{"en":395,"de":396,"sr":397,"es":398,"fr":399,"it":400,"ru":401,"zh":402},"\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fde\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fsr\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fes\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Ffr\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fit\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fru\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning","\u002Fzh\u002Fblog\u002Fqwen-3-6-in-production-release-runbook-ai-rollback-and-llmops-versioning",[404,408,412,416,420],{"id":405,"name":406,"slug":407},39,"Enterprise Delivery OS","enterprise",{"id":409,"name":410,"slug":411},87,"Playbook: LLMOps","llmops",{"id":413,"name":414,"slug":415},88,"Verzionisanje (prompt, modeli)","versioning",{"id":417,"name":418,"slug":419},94,"Runbook: Izdanje","release-runbook",{"id":421,"name":422,"slug":423},106,"Runbook: AI rollback","ai-rollback",{"id":425,"login":426,"email":427,"displayName":428},"20","rooth8233","aleksandar@stajic.de","Aleksandar Stajić",[430,600],{"lang":431,"title":432,"content":433,"contentJson":434,"excerpt":599},"en","Qwen 3.6 in Production: Release Runbook, AI Rollback, and LLMOps Versioning","{\"time\":1774828800000,\"blocks\":[{\"data\":{\"text\":\"\u003Cb>Qwen3.6-Plus\u003C\u002Fb> matters because it pushes the Qwen line from a promising agentic model into something much closer to a production-grade execution layer. The earlier Qwen 3.5-Plus article on stajic.de already framed the shift correctly: the market is moving away from chat-only intelligence and toward reliable multi-step execution. Qwen3.6 takes that direction further with stronger agentic coding, better multimodal reasoning, and a more stability-focused release posture.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"That makes this topic a natural fit for \u003Cb>Enterprise Delivery OS\u003C\u002Fb>. It belongs primarily under \u003Cb>LLMOps Playbook\u003C\u002Fb>, with the strongest sub-fit under \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. At the same time, it should also sit naturally inside \u003Cb>Release Runbook\u003C\u002Fb> and \u003Cb>AI Rollback Runbook\u003C\u002Fb>, because a model upgrade is not just a model choice. It is a release event, a rollback scenario, and a versioning problem at the same time.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"The official Qwen3.6-Plus launch positions the model as a major upgrade over Qwen3.5-Plus, especially for agentic coding, repository-level problem solving, multimodal reasoning, and stable real-world execution. Alibaba also states that the hosted Plus model is available immediately with a 1M context window by default, while open-weight Qwen3.6 variants extend the family for teams that want more control over deployment and inference choices.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"What Changed from Qwen 3.5 to Qwen 3.6\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The original Qwen 3.5-Plus article on stajic.de focused on four practical strengths: large context, tool-use behavior, multimodal capability, and the move toward reliable agentic execution. Qwen3.6-Plus keeps that foundation, but the official release sharpens the operational value. It puts much more emphasis on agentic coding quality, terminal-style execution, long-horizon tool use, and stronger stability based on deployment feedback from the Qwen3.5 era.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"1M context window by default in the hosted Plus model\",\"Significantly improved agentic coding capability\",\"Better multimodal perception and reasoning\",\"A more stable and reliable base for real-world developer workflows\",\"A broader Qwen3.6 family that also includes open-weight variants for self-hosted use cases\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"code\":\"\u002F\u002F Minimal migration idea\\nconst modelConfig = { provider: \\\"qwen\\\", model: \\\"qwen3.6-plus\\\", maxContext: 1000000, mode: \\\"agentic-coding\\\", tools: [\\\"browser\\\", \\\"bash\\\", \\\"search\\\", \\\"file-edit\\\"]\\n}; \u002F\u002F The hard part is not the model switch.\\n\u002F\u002F The hard part is release control, evaluation, and rollback readiness.\"},\"type\":\"code\"},{\"data\":{\"text\":\"Why Qwen 3.6 Is a Release Runbook Topic\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The live \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\\\">Release Runbook\u003C\u002Fa> defines release safety through preflight checks, clear owners, verification against acceptance criteria, captured evidence, and post-release review. A production upgrade from Qwen3.5-Plus to Qwen3.6-Plus fits that pattern exactly. A model release is not just a new feature. It is a behavior change inside a live system, and that means it deserves release-grade discipline.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"This becomes even more important when the model is used for code generation, tool execution, repo-level reasoning, or multimodal workflows. The larger the operational surface, the more dangerous it becomes to treat the upgrade as a single configuration tweak.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"Release checklist for a model upgrade\\n1. Define the target version and deployment scope\\n2. Freeze prompt and routing changes during validation\\n3. Run the evaluation harness on a stable test set\\n4. Verify cost, latency, and failure-rate deltas\\n5. Confirm rollback path and rollback trigger thresholds\\n6. Approve release with named owner and evidence pack\\n7. Monitor post-release behavior before full traffic shift\"},\"type\":\"code\"},{\"data\":{\"text\":\"Why Qwen 3.6 Is Also an AI Rollback Topic\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The live \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Fai-rollback\\\">AI Rollback Runbook\u003C\u002Fa> is explicit: LLM systems can regress through prompt changes, routing changes, model updates, or data drift. That is not theoretical. A model upgrade can improve coding benchmarks and still regress a production workflow that depends on formatting stability, tool discipline, safety behavior, cost profile, or output style.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Qwen3.6-Plus may be stronger overall, but production systems do not fail on average quality. They fail on edge behavior, hidden dependencies, and brittle integration assumptions. That is why every model upgrade needs explicit rollback conditions before traffic moves.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"{ \\\"rollbackTriggers\\\": { \\\"qualityDropPct\\\": 5, \\\"toolCallFailurePct\\\": 2, \\\"costIncreasePct\\\": 20, \\\"latencyIncreasePct\\\": 25, \\\"safetyViolationCount\\\": 1 }, \\\"rollbackTarget\\\": \\\"qwen3.5-plus\\\", \\\"freezeWindow\\\": \\\"24h\\\", \\\"requiredEvidence\\\": [ \\\"eval-report\\\", \\\"traffic-split-report\\\", \\\"post-release-verification\\\" ]\\n}\"},\"type\":\"code\"},{\"data\":{\"text\":\"Versioning Prompts and Models Is the Control Surface\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The live \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\\\">Versioning (Prompts, Models)\u003C\u002Fa> page gives the right framing: prompts and models need traceability and controlled change. That point becomes much more concrete when a model family upgrades quickly. If the team cannot say which prompt set, routing logic, temperature policy, tool permissions, and evaluation baseline were active at a given release, then the system is not truly versioned. It is merely configured.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"A production-grade version record should bind model identity, prompt bundle, tool policy, evaluation set, and release decision together. This is especially important for Qwen3.6 because the model is designed for stronger agentic execution. More capable behavior means a bigger need for explicit version boundaries.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"{ \\\"versionId\\\": \\\"llm-stack-2026-05-04-a\\\", \\\"model\\\": \\\"qwen3.6-plus\\\", \\\"fallbackModel\\\": \\\"qwen3.5-plus\\\", \\\"promptBundle\\\": \\\"repo-agent-v12\\\", \\\"toolPolicy\\\": \\\"repo-agent-safe-tools-v4\\\", \\\"routerPolicy\\\": \\\"coding-heavy-workloads-v3\\\", \\\"evalSet\\\": \\\"agentic-coding-regression-suite-v7\\\", \\\"approvedBy\\\": \\\"llmops-owner\\\", \\\"releaseState\\\": \\\"canary\\\"\\n}\"},\"type\":\"code\"},{\"data\":{\"text\":\"Why the Best Primary Placement Is LLMOps Playbook\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The live \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\\\">LLMOps Playbook\u003C\u002Fa> already defines the key operating logic: version prompts and models, evaluate with quality gates, roll out through canary or A\u002FB paths, monitor for drift or regressions, and keep rollback fast. Qwen3.6-Plus is almost a textbook example of why that playbook exists. The model is stronger, but the upgrade only becomes valuable when behavior stays stable across changes.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"Versioning protects traceability and controlled change\",\"Evaluation harness protects quality before traffic shift\",\"Canary and A\u002FB releases reduce model-upgrade blast radius\",\"Monitoring catches regressions that static evaluation missed\",\"Rollback strategy keeps the system reversible when real traffic exposes weaknesses\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"This is also why the live \u003Ca href=\\\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fcanary-and-ab\\\">Canary and A\u002FB Releases\u003C\u002Fa> page fits naturally here. A capable model should not go directly from benchmark excitement to full production traffic. The safer pattern is staged rollout with explicit evidence at each stage.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"LLMOps rollout sequence\\n1. Pin the exact model and prompt bundle\\n2. Run offline evaluation on a fixed regression suite\\n3. Start canary traffic with explicit guardrails\\n4. Compare quality, latency, cost, and tool-failure metrics\\n5. Expand traffic only if thresholds hold\\n6. Keep rollback immediate and documented\"},\"type\":\"code\"},{\"data\":{\"text\":\"Where Qwen 3.6 Creates Real Leverage\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The strongest practical value of Qwen3.6 is not that it sounds smarter in a demo. The real leverage comes where workflow continuity matters: repository-level coding, long-horizon tool use, multimodal debugging, and repeated execution under changing context. That is where a more agentic model can remove real friction instead of just improving headline benchmarks.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"Repo-level coding tasks with multiple files and longer dependency chains\",\"Terminal-oriented execution paths where tool discipline matters\",\"Multimodal QA and UI debugging where screenshots and documents matter\",\"Ops and incident analysis with larger context and runbook-style execution\",\"Agentic workflows where stability across many steps matters more than single-answer brilliance\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"That direction is consistent with both the Qwen3.6-Plus launch and the open-weight Qwen3.6 releases, which emphasize agentic coding, repository reasoning, thinking preservation, and broader deployment flexibility. For teams already testing Qwen3.5-Plus, the question is no longer whether Qwen3.6 is interesting. The real question is whether the team can upgrade it with the same discipline used for any other production dependency.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Trade-offs to Respect Before Upgrading\",\"level\":2},\"type\":\"header\"},{\"data\":{\"items\":[\"A larger context window does not remove the need for structured inputs and retrieval planning\",\"Stronger agentic coding increases the need for tool policy, sandboxing, and replayable logs\",\"Hosted and open-weight variants create different release, privacy, and operations trade-offs\",\"A better average model can still regress one critical workflow if evaluation is weak\",\"Version drift across prompts, routers, and model endpoints can destroy traceability if left uncontrolled\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"A model upgrade is only a capability upgrade if the operating system around it can prove stability, trace the change, and reverse it fast.\",\"caption\":\"LLMOps perspective\"},\"type\":\"quote\"},{\"data\":{\"text\":\"Best-Fit Pillar Placement\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The best primary placement for this article is \u003Cb>LLMOps Playbook\u003C\u002Fb>, with the strongest sub-fit under \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. Additional placement under \u003Cb>Release Runbook\u003C\u002Fb> and \u003Cb>AI Rollback Runbook\u003C\u002Fb> is justified because the topic is explicitly about controlled release, rollback readiness, and model-version traceability. In other words, Qwen3.6 is not only a model story. It is an operations story.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Final Perspective\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Qwen3.6-Plus is a strong signal that agentic AI is becoming more practical for serious engineering work. But the real maturity signal is not the benchmark graph. It is whether a team can release the new model through an evidence-backed runbook, preserve prompt and model version traceability, canary the change safely, monitor the real behavior, and roll back quickly if one critical workflow regresses. That is the difference between experimenting with models and operating them.\"},\"type\":\"paragraph\"},{\"data\":{},\"type\":\"delimiter\"},{\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fnew-qwen-3-5-plus\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"New Qwen 3.5-Plus: Open-source AI is getting serious now\",\"description\":\"Earlier article on Qwen 3.5-Plus as the shift from chat-style intelligence toward more reliable agentic execution.\"}},\"type\":\"linkTool\"},{\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"LLMOps Playbook\",\"description\":\"Keep LLM behavior stable across changes through versioning, evaluation, canary releases, monitoring, and fast rollback procedures.\"}},\"type\":\"linkTool\"},{\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Versioning (Prompts, Models)\",\"description\":\"Versioning strategy for prompts and models to ensure traceability and controlled change.\"}},\"type\":\"linkTool\"},{\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Release Runbook\",\"description\":\"Use preflight checks, named owners, verification against acceptance criteria, captured evidence, and post-release review.\"}},\"type\":\"linkTool\"},{\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Fai-rollback\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"AI Rollback Runbook\",\"description\":\"LLM systems can regress through prompt changes, routing changes, model updates, or data drift. Freeze, verify, rollback, and learn.\"}},\"type\":\"linkTool\"}],\"version\":\"2.30.8\"}",{"time":435,"blocks":436,"version":598},1774828800000,[437,440,443,446,449,452,460,463,466,469,472,475,478,481,484,486,489,492,495,497,500,503,511,514,517,520,523,531,534,537,545,549,552,555,558,561,563,570,577,584,591],{"data":438,"type":217},{"text":439},"\u003Cb>Qwen3.6-Plus\u003C\u002Fb> matters because it pushes the Qwen line from a promising agentic model into something much closer to a production-grade execution layer. The earlier Qwen 3.5-Plus article on stajic.de already framed the shift correctly: the market is moving away from chat-only intelligence and toward reliable multi-step execution. Qwen3.6 takes that direction further with stronger agentic coding, better multimodal reasoning, and a more stability-focused release posture.",{"data":441,"type":217},{"text":442},"That makes this topic a natural fit for \u003Cb>Enterprise Delivery OS\u003C\u002Fb>. It belongs primarily under \u003Cb>LLMOps Playbook\u003C\u002Fb>, with the strongest sub-fit under \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. At the same time, it should also sit naturally inside \u003Cb>Release Runbook\u003C\u002Fb> and \u003Cb>AI Rollback Runbook\u003C\u002Fb>, because a model upgrade is not just a model choice. It is a release event, a rollback scenario, and a versioning problem at the same time.",{"data":444,"type":217},{"text":445},"The official Qwen3.6-Plus launch positions the model as a major upgrade over Qwen3.5-Plus, especially for agentic coding, repository-level problem solving, multimodal reasoning, and stable real-world execution. Alibaba also states that the hosted Plus model is available immediately with a 1M context window by default, while open-weight Qwen3.6 variants extend the family for teams that want more control over deployment and inference choices.",{"data":447,"type":42},{"text":448,"level":227},"What Changed from Qwen 3.5 to Qwen 3.6",{"data":450,"type":217},{"text":451},"The original Qwen 3.5-Plus article on stajic.de focused on four practical strengths: large context, tool-use behavior, multimodal capability, and the move toward reliable agentic execution. Qwen3.6-Plus keeps that foundation, but the official release sharpens the operational value. It puts much more emphasis on agentic coding quality, terminal-style execution, long-horizon tool use, and stronger stability based on deployment feedback from the Qwen3.5 era.",{"data":453,"type":240},{"items":454,"style":239},[455,456,457,458,459],"1M context window by default in the hosted Plus model","Significantly improved agentic coding capability","Better multimodal perception and reasoning","A more stable and reliable base for real-world developer workflows","A broader Qwen3.6 family that also includes open-weight variants for self-hosted use cases",{"data":461,"type":244},{"code":462},"\u002F\u002F Minimal migration idea\nconst modelConfig = { provider: \"qwen\", model: \"qwen3.6-plus\", maxContext: 1000000, mode: \"agentic-coding\", tools: [\"browser\", \"bash\", \"search\", \"file-edit\"]\n}; \u002F\u002F The hard part is not the model switch.\n\u002F\u002F The hard part is release control, evaluation, and rollback readiness.",{"data":464,"type":42},{"text":465,"level":227},"Why Qwen 3.6 Is a Release Runbook Topic",{"data":467,"type":217},{"text":468},"The live \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Frelease-runbook\">Release Runbook\u003C\u002Fa> defines release safety through preflight checks, clear owners, verification against acceptance criteria, captured evidence, and post-release review. A production upgrade from Qwen3.5-Plus to Qwen3.6-Plus fits that pattern exactly. A model release is not just a new feature. It is a behavior change inside a live system, and that means it deserves release-grade discipline.",{"data":470,"type":217},{"text":471},"This becomes even more important when the model is used for code generation, tool execution, repo-level reasoning, or multimodal workflows. The larger the operational surface, the more dangerous it becomes to treat the upgrade as a single configuration tweak.",{"data":473,"type":244},{"code":474},"Release checklist for a model upgrade\n1. Define the target version and deployment scope\n2. Freeze prompt and routing changes during validation\n3. Run the evaluation harness on a stable test set\n4. Verify cost, latency, and failure-rate deltas\n5. Confirm rollback path and rollback trigger thresholds\n6. Approve release with named owner and evidence pack\n7. Monitor post-release behavior before full traffic shift",{"data":476,"type":42},{"text":477,"level":227},"Why Qwen 3.6 Is Also an AI Rollback Topic",{"data":479,"type":217},{"text":480},"The live \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Fai-rollback\">AI Rollback Runbook\u003C\u002Fa> is explicit: LLM systems can regress through prompt changes, routing changes, model updates, or data drift. That is not theoretical. A model upgrade can improve coding benchmarks and still regress a production workflow that depends on formatting stability, tool discipline, safety behavior, cost profile, or output style.",{"data":482,"type":217},{"text":483},"Qwen3.6-Plus may be stronger overall, but production systems do not fail on average quality. They fail on edge behavior, hidden dependencies, and brittle integration assumptions. That is why every model upgrade needs explicit rollback conditions before traffic moves.",{"data":485,"type":244},{"code":268},{"data":487,"type":42},{"text":488,"level":227},"Versioning Prompts and Models Is the Control Surface",{"data":490,"type":217},{"text":491},"The live \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning\">Versioning (Prompts, Models)\u003C\u002Fa> page gives the right framing: prompts and models need traceability and controlled change. That point becomes much more concrete when a model family upgrades quickly. If the team cannot say which prompt set, routing logic, temperature policy, tool permissions, and evaluation baseline were active at a given release, then the system is not truly versioned. It is merely configured.",{"data":493,"type":217},{"text":494},"A production-grade version record should bind model identity, prompt bundle, tool policy, evaluation set, and release decision together. This is especially important for Qwen3.6 because the model is designed for stronger agentic execution. More capable behavior means a bigger need for explicit version boundaries.",{"data":496,"type":244},{"code":280},{"data":498,"type":42},{"text":499,"level":227},"Why the Best Primary Placement Is LLMOps Playbook",{"data":501,"type":217},{"text":502},"The live \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\">LLMOps Playbook\u003C\u002Fa> already defines the key operating logic: version prompts and models, evaluate with quality gates, roll out through canary or A\u002FB paths, monitor for drift or regressions, and keep rollback fast. Qwen3.6-Plus is almost a textbook example of why that playbook exists. The model is stronger, but the upgrade only becomes valuable when behavior stays stable across changes.",{"data":504,"type":240},{"items":505,"style":239},[506,507,508,509,510],"Versioning protects traceability and controlled change","Evaluation harness protects quality before traffic shift","Canary and A\u002FB releases reduce model-upgrade blast radius","Monitoring catches regressions that static evaluation missed","Rollback strategy keeps the system reversible when real traffic exposes weaknesses",{"data":512,"type":217},{"text":513},"This is also why the live \u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fcanary-and-ab\">Canary and A\u002FB Releases\u003C\u002Fa> page fits naturally here. A capable model should not go directly from benchmark excitement to full production traffic. The safer pattern is staged rollout with explicit evidence at each stage.",{"data":515,"type":244},{"code":516},"LLMOps rollout sequence\n1. Pin the exact model and prompt bundle\n2. Run offline evaluation on a fixed regression suite\n3. Start canary traffic with explicit guardrails\n4. Compare quality, latency, cost, and tool-failure metrics\n5. Expand traffic only if thresholds hold\n6. Keep rollback immediate and documented",{"data":518,"type":42},{"text":519,"level":227},"Where Qwen 3.6 Creates Real Leverage",{"data":521,"type":217},{"text":522},"The strongest practical value of Qwen3.6 is not that it sounds smarter in a demo. The real leverage comes where workflow continuity matters: repository-level coding, long-horizon tool use, multimodal debugging, and repeated execution under changing context. That is where a more agentic model can remove real friction instead of just improving headline benchmarks.",{"data":524,"type":240},{"items":525,"style":239},[526,527,528,529,530],"Repo-level coding tasks with multiple files and longer dependency chains","Terminal-oriented execution paths where tool discipline matters","Multimodal QA and UI debugging where screenshots and documents matter","Ops and incident analysis with larger context and runbook-style execution","Agentic workflows where stability across many steps matters more than single-answer brilliance",{"data":532,"type":217},{"text":533},"That direction is consistent with both the Qwen3.6-Plus launch and the open-weight Qwen3.6 releases, which emphasize agentic coding, repository reasoning, thinking preservation, and broader deployment flexibility. For teams already testing Qwen3.5-Plus, the question is no longer whether Qwen3.6 is interesting. The real question is whether the team can upgrade it with the same discipline used for any other production dependency.",{"data":535,"type":42},{"text":536,"level":227},"Trade-offs to Respect Before Upgrading",{"data":538,"type":240},{"items":539,"style":239},[540,541,542,543,544],"A larger context window does not remove the need for structured inputs and retrieval planning","Stronger agentic coding increases the need for tool policy, sandboxing, and replayable logs","Hosted and open-weight variants create different release, privacy, and operations trade-offs","A better average model can still regress one critical workflow if evaluation is weak","Version drift across prompts, routers, and model endpoints can destroy traceability if left uncontrolled",{"data":546,"type":333},{"text":547,"caption":548},"A model upgrade is only a capability upgrade if the operating system around it can prove stability, trace the change, and reverse it fast.","LLMOps perspective",{"data":550,"type":42},{"text":551,"level":227},"Best-Fit Pillar Placement",{"data":553,"type":217},{"text":554},"The best primary placement for this article is \u003Cb>LLMOps Playbook\u003C\u002Fb>, with the strongest sub-fit under \u003Cb>Versioning (Prompts, Models)\u003C\u002Fb>. Additional placement under \u003Cb>Release Runbook\u003C\u002Fb> and \u003Cb>AI Rollback Runbook\u003C\u002Fb> is justified because the topic is explicitly about controlled release, rollback readiness, and model-version traceability. In other words, Qwen3.6 is not only a model story. It is an operations story.",{"data":556,"type":42},{"text":557,"level":227},"Final Perspective",{"data":559,"type":217},{"text":560},"Qwen3.6-Plus is a strong signal that agentic AI is becoming more practical for serious engineering work. But the real maturity signal is not the benchmark graph. It is whether a team can release the new model through an evidence-backed runbook, preserve prompt and model version traceability, canary the change safely, monitor the real behavior, and roll back quickly if one critical workflow regresses. That is the difference between experimenting with models and operating them.",{"data":562,"type":348},{},{"data":564,"type":357},{"link":565,"meta":566},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fnew-qwen-3-5-plus",{"image":567,"title":568,"description":569},{"url":354},"New Qwen 3.5-Plus: Open-source AI is getting serious now","Earlier article on Qwen 3.5-Plus as the shift from chat-style intelligence toward more reliable agentic execution.",{"data":571,"type":357},{"link":572,"meta":573},"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops",{"image":574,"title":575,"description":576},{"url":354},"LLMOps Playbook","Keep LLM behavior stable across changes through versioning, evaluation, canary releases, monitoring, and fast rollback procedures.",{"data":578,"type":357},{"link":579,"meta":580},"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Fplaybooks\u002Fllmops\u002Fversioning",{"image":581,"title":582,"description":583},{"url":354},"Versioning (Prompts, Models)","Versioning strategy for prompts and models to ensure traceability and controlled change.",{"data":585,"type":357},{"link":586,"meta":587},"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Frelease-runbook",{"image":588,"title":589,"description":590},{"url":354},"Release Runbook","Use preflight checks, named owners, verification against acceptance criteria, captured evidence, and post-release review.",{"data":592,"type":357},{"link":593,"meta":594},"https:\u002F\u002Fstajic.de\u002Fenterprise\u002Frunbooks\u002Fai-rollback",{"image":595,"title":596,"description":597},{"url":354},"AI Rollback Runbook","LLM systems can regress through prompt changes, routing changes, model updates, or data drift. Freeze, verify, rollback, and learn.","2.30.8","Qwen 3.6 is not just another model upgrade. It is a release event, a rollback scenario, and a versioning problem at the same time. This article explains how Qwen 3.6 should be handled in production through LLMOps discipline, prompt and model traceability, controlled rollout, and evidence-based rollback readiness.",{"lang":7,"title":208,"content":210,"contentJson":601,"excerpt":387},{"time":212,"blocks":602,"version":386},[603,605,607,609,611,613,616,618,620,622,624,626,628,630,632,634,636,638,640,642,644,646,649,651,653,655,657,660,662,664,667,669,671,673,675,677,679,683,687,691,695],{"data":604,"type":217},{"text":216},{"data":606,"type":217},{"text":220},{"data":608,"type":217},{"text":223},{"data":610,"type":42},{"text":226,"level":227},{"data":612,"type":217},{"text":230},{"data":614,"type":240},{"items":615,"style":239},[234,235,236,237,238],{"data":617,"type":244},{"code":243},{"data":619,"type":42},{"text":247,"level":227},{"data":621,"type":217},{"text":250},{"data":623,"type":217},{"text":253},{"data":625,"type":244},{"code":256},{"data":627,"type":42},{"text":259,"level":227},{"data":629,"type":217},{"text":262},{"data":631,"type":217},{"text":265},{"data":633,"type":244},{"code":268},{"data":635,"type":42},{"text":271,"level":227},{"data":637,"type":217},{"text":274},{"data":639,"type":217},{"text":277},{"data":641,"type":244},{"code":280},{"data":643,"type":42},{"text":283,"level":227},{"data":645,"type":217},{"text":286},{"data":647,"type":240},{"items":648,"style":239},[290,291,292,293,294],{"data":650,"type":217},{"text":297},{"data":652,"type":244},{"code":300},{"data":654,"type":42},{"text":303,"level":227},{"data":656,"type":217},{"text":306},{"data":658,"type":240},{"items":659,"style":239},[310,311,312,313,314],{"data":661,"type":217},{"text":317},{"data":663,"type":42},{"text":320,"level":227},{"data":665,"type":240},{"items":666,"style":239},[324,325,326,327,328],{"data":668,"type":333},{"text":331,"caption":332},{"data":670,"type":42},{"text":336,"level":227},{"data":672,"type":217},{"text":339},{"data":674,"type":42},{"text":342,"level":227},{"data":676,"type":217},{"text":345},{"data":678,"type":348},{},{"data":680,"type":357},{"link":351,"meta":681},{"image":682,"title":355,"description":356},{"url":354},{"data":684,"type":357},{"link":360,"meta":685},{"image":686,"title":363,"description":364},{"url":354},{"data":688,"type":357},{"link":367,"meta":689},{"image":690,"title":370,"description":371},{"url":354},{"data":692,"type":357},{"link":374,"meta":693},{"image":694,"title":377,"description":378},{"url":354},{"data":696,"type":357},{"link":381,"meta":697},{"image":698,"title":384,"description":385},{"url":354},"Post erfolgreich abgerufen",{"items":701,"source":708,"manualIds":709,"manualMatchedIds":711},[702],{"id":703,"slug":704,"title":705,"excerpt":706,"featuredImage":388,"publishedAt":707},"384","new-qwen-3-5-plus","Novi Qwen 3.5-Plus: AI otvorenog koda je upravo postao ozbiljan.","Otkrijte revolucionarne funkcije i prednosti Alibabinog Qwen 3.5-Plus modela, AI otvorenog koda koji menja pravila igre za programere.","2026-02-19T10:23:00.000Z","manual",[710],384,[710]]