Kwenzekeni
U-Alibaba ukhiphe i-Qwen3.8-Omni-Flash, imodeli entsha ochungechungeni lwayo lwe-Qwen eklanyelwe ukucutshungulwa kwe-omnimodal. Imodeli isekela iwindi lomongo wamathokheni ayisigidi futhi iqondise ukuhamba komsebenzi njengokuhlela ividiyo, ukukhiqizwa kwamavidiyo omculo, nengxoxo yesikhathi sangempela. Ngokusho kwe-Gigazine, imodeli idlula eyandulelayo, i-Qwen3.5-Omni-Plus, ngaphezu kuka-25% ngokwesilinganiso kuwo wonke ama-benchmarks angu-29. U-Alibaba uthi imodeli entsha izuza ukusebenza komsindo nevidiyo eduze noma kwedlula i-Gemini 3.8 Flash, kuyilapho yehlisa izindleko ze-API zokufakwa kwezwi ngaphezu kuka-98% kanye nokokufaka kwezwi/kwevidiyo ngaphezu kuka-93% uma kuqhathaniswa nenguqulo yangaphambilini. Ukukhishwa kufaka phakathi izandiso kuma-Qwen-MM-Plugins kanye nehhanisi elivula umthombo elibizwa ngokuthi i-Qwen-Live Harness, nakuba leli khasi lakamuva le-GitHub lalingenakufinyeleleka ngesikhathi sokubika.
I-Alibaba yengeze i-Qwen3.8-Omni-Flash ochungechungeni lwayo lwe-Qwen, iluchaza njengemodeli yendabuko ye-omnimodal yesizukulwane esilandelayo. Imodeli yakhelwe ukuphatha izinhlobonhlobo zokugeleza komsebenzi, okuhlanganisa ukuhlela ividiyo, ukukhiqizwa kwevidiyo yomculo, ukukhiqizwa kwefilimu, ukufingqa okulalelwayo nokubonwayo, kanye nengxoxo yesikhathi sangempela. Isekela iwindi lomongo lamathokheni ayizigidi ezingu-1, okuwukukhuphuka okukhulu okuvumela ukucutshungulwa kwamafayela omsindo nevidiyo efomu elide.
Ngokusho kwe-Gigazine, imodeli iletha imiphumela ephakeme kakhulu uma iqhathaniswa neyandulelayo, i-Qwen3.5-Omni-Plus. Kuma-benchmarks angama-29, isilinganiso samaphuzu sibikwa singaphezulu kuka-25%. Ukuthuthukiswa okuqondile kuyaphawulwa emisebenzini ewumongo njengokuqonda umsindo omude, okucatshangelwayo komsindo/ividiyo, amagama-ncazo, nokubonwa kwezipikha eziningi. Isibonelo, ithole amaphuzu angu-82.7 ku-LongAudioSpan kanye nama-89.7 ku-AliMeeting, ibhentshimark yokubhala umsindo wenkomfa yeziteshi eziningi zabantu abaningi ngesiShayina.
U-Alibaba uthi i-Qwen3.8-Omni-Flash ikala idatha, umongo, kanye nezindawo ze-ejenti ukuze kuzuzwe ukusebenza komsindo nevidiyo eduze naleyo ye-Gemini 3.8 Flash, nokusebenza komsindo kukonke okuyedlulayo. Inkampani igqamisa ukwehliswa kwezindleko okubalulekile, isho ukuthi izindleko ze-API ngehora zokufakwa kwezwi zingaphansi kwama-98%, futhi ngokokufaka kwezwi nevidiyo, zingaphansi kuka-93% kunemodeli yangaphambilini. Lezi zimangalo zezindleko ziyingxenye yesiphakamiso senani lemodeli kubasebenzisi bebhizinisi.
Ukuze kusekelwe amakhono emodeli, i-Alibaba inwebise i-Qwen-MM-Plugins ukuze yengeze umbono lapho kudingeka khona, ukusetshenziswa kwamathuluzi, namandla okusebenzisa ukugeleza komsebenzi omsindo nevidiyo ende. Le nkampani iphinde yamemezela ukutholakala okuvulekile kwe-Qwen-Live Harness, ihhanisi elibanzi eliklanywe ngokusekelwe ku-Qwen3.8-Omni-Flash-Realtime API. Kodwa-ke, i-Gigazine iphawula ukuthi ngesikhathi sokubhala, ikhasi le-Qwen-Live Harness GitHub belingafinyeleleki futhi libuyisele iphutha le-404, elibonisa ukubambezeleka okungenzeka noma izinkinga ngokukhishwa komphakathi kwale ngxenye.
Imininingwane yomthombo: gigazine.net ↗
Kungani kubalulekile
Lokhu kukhishwa kubalulekile ngoba kubheka izindleko eziphezulu nobunzima bokucubungula idatha yefomu elide lomsindo nevidiyo kuma-agent we-AI. Ngokufuna ukusebenza komngcele ngengxenyana yezindleko zamamodeli angaphambilini, i-Alibaba ibeka i-Qwen3.8-Omni-Flash njengethuluzi elisebenzayo lezinhlelo zokusebenza zebhizinisi ezidinga ukucabanga kwesikhathi sangempela kwezindlela eziningi. Ukushintsha kusukela ekuphatheni okulalelwayo nevidiyo njengokufakwayo okulula kokuqonda ukuya kumidiya eyinhloko yokucabanga komenzeli kungasheshisa ukuhlanganiswa kwe-AI kuzimo zokukhiqiza zomhlaba wangempela, njengokukhiqizwa kwemidiya okuzenzakalelayo kanye nama-ejenti ezingxoxo ayinkimbinkimbi. Nokho, ukuqhathaniswa kwebhentshimakhi ethile ne-Gemini 3.8 Flash kusekelwe ezimangalweni ze-Alibaba futhi akuzange kuqinisekiswe ngokuzimela abahloli bezinkampani zangaphandle.
Ukukhululwa kwe-Qwen3.8-Omni-Flash kubalulekile embonini ye-AI ngoba iqondise iphuzu elithile lobuhlungu: izindleko eziphezulu kanye yinkimbinkimbi yobuchwepheshe yokucubungula idatha ye-multimodal yefomu elide. Ngokufuna ukusebenza komngcele ngezindleko ezehliswe kakhulu, i-Alibaba iphonsela inselelo isimo esikhona sentengo ye-AI ye-multimodal nokufinyeleleka. Lokhu kungaholela ekwamukelweni okubanzi kwama-ejenti e-AI ezimbonini ezithembele kakhulu kudatha yomsindo nevidiyo, njengokukhiqizwa kwemidiya, isevisi yamakhasimende, nokuhumusha kwesikhathi sangempela.
Ikhono lemodeli lokuphatha amathokheni womongo angu-1 million liyintuthuko enkulu yobuchwepheshe, njengoba livumela ukucutshungulwa kwamafayela omsindo namavidyo amade ngaphandle kokuhlukaniswa. Lokhu kubalulekile ezinhlelweni ezifana nokukhiqizwa kwefilimu noma ukulotshwa komhlangano wefomu elide, lapho ukuqhubeka komongo kubalulekile. Ukuthuthukiswa kokuqashelwa kwezikhulumi eziningi kanye nokuqonda komsindo omude kuphakamisa ukuthi imodeli ifaneleka kangcono izimo eziyinkimbinkimbi, zomhlaba wangempela kuneziphindaphindo zangaphambilini.
Kodwa-ke, izimangalo zokusebenza okungaphezu kokusebenza kahle noma okufanayo kwe-Gemini 3.8 Flash zisekelwe kumabhentshimakhi angaphakathi we-Alibaba futhi azikakaqinisekiswanga ngokuzimela. Le yinkinga evamile embonini ye-AI, lapho izinkampani zivame ukuthembela kumamethrikhi azibikayo. Ukuhlola okuzimele kuyodingeka ukuze kuqinisekiswe ukusebenza kweqiniso kanye nokuphumelela kwezindleko zemodeli. Kuze kube yileso sikhathi, amabhizinisi kufanele akhulume nezimangalo ngokuqapha futhi azenzele ezawo izivivinyo ngaphambi kokutshala izimali ezibalulekile.
Ukunikezwa okuvulekile kwe-Qwen-Live Harness kuyisinyathelo esihle somphakathi wonjiniyela, njengoba kunikeza uhlaka lokwakha izinhlelo zokusebenza phezu kwemodeli. Nokho, ukungafinyeleleki kwekhasi le-GitHub ngesikhathi sokubika kuyifulegi elibomvu elingabambezela ukutholwa. Onjiniyela bazodinga ukuthi balinde ukuthi ihhanisi litholakale ngokugcwele futhi lizinze ngaphambi kokuba baqale ukwakha izinhlelo zokusebenza ezilungele ukukhiqiza.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Onjiniyela kufanele baqaphe ukutholakala nokuzinza kwe-Qwen-Live Harness, njengoba umthombo uphawula ukuthi ikhasi lawo le-GitHub libuyise iphutha le-404. Ukwengeza, amabhentshimakhi azimele aqhathanisa i-Qwen3.8-Omni-Flash namanye amamodeli asemngceleni afana ne-Gemini 3.8 Flash azobaluleka ukuze kuqinisekiswe izimangalo zokusebenza kwe-Alibaba. Umthelela ongokoqobo entengo ye-API yemisebenzi ye-multimodal futhi uzoba isici esibalulekile kumabhizinisi acabangela ukutholwa.
Ukutholakala nokuzinza kwe-Qwen-Live Harness kuzoba yisici esibalulekile ekuthathweni kwemodeli. Uma ikhasi le-GitHub lihlala lingafinyeleleki noma uma ihhanisi lineziphazamisi ezibalulekile, lingavimbela onjiniyela ekwakheni izinhlelo zokusebenza ngaphezulu kwe-Qwen3.8-Omni-Flash. Ukuqapha inqolobane ye-GitHub nanoma yiziphi izibuyekezo ezivela e-Alibaba zizobalulekile.
Amabhentshimakhi azimele aqhathanisa i-Qwen3.8-Omni-Flash namanye amamodeli asemngceleni, njenge-Gemini 3.8 Flash kanye ne-GPT-4o, azobaluleka kakhulu ukuze kuqinisekiswe izimangalo zokusebenza kwe-Alibaba. Ukuhlola okuvela eceleni kuzonikeza umbono onenjongo kakhudlwana wamakhono emodeli futhi kusize amabhizinisi ukwenza izinqumo ezinolwazi mayelana nokutholwa.
Umthelela obonakalayo entengo ye-API yemisebenzi ye-multimodal futhi uzoba isici esibalulekile. Uma ukuncishiswa kwezindleko ezifunwa yi-Alibaba kwenzeka, kungaholela ekushintsheni okukhulu emakethe, okwenza i-multimodal AI ifinyeleleke kalula ezinkampanini ezincane nabathuthukisi. Ukuqapha izindleko zangempela ze-API nokuziqhathanisa nezimbangi kuzobaluleka.
Ukusebenza kwemodeli ezimeni zomhlaba wangempela, njengokuhlela ividiyo kanye nengxoxo yesikhathi sangempela, kuzoba inkomba ebalulekile yokusetshenziswa kwayo okungokoqobo. Impendulo yabasebenzisi kanye nezifundo eziyisibonelo ezivela kubamukeli bokuqala zizohlinzeka ngemininingwane ebalulekile yamandla nemikhawulo yemodeli.