I-Google ithathe isinyathelo enkundleni yobuhlakani bokwenziwa. Le nkampani iveze i-Gemini 4 Argon , imodeli yayo entsha esezingeni eliphezulu ehlose ukuncintisana ngayo nezinhlelo ezinamandla kakhulu ezitholakalayo njengamanje. Lesi simemezelo siza ngemuva kwezinyanga ezimbalwa lapho inkampani ibonakala sengathi ilahlekelwe yithuba lokuthuthuka kwe-OpenAI kanye ne-Anthropic.
Imodeli yenzelwe imisebenzi yobungcweti eyinkimbinkimbi , njengohlelo, ukuphepha kwe-cyber, noma umsebenzi wolwazi lwebhizinisi. Ngokusho kwedatha eyabiwe yi-Google, i-Argon ifanisa noma idlula ngisho nezilinganiso ezincintisanayo ezivivinyweni eziningana, yize imibono ngaphakathi nangaphandle kwenkampani ingavumelani ngokuphelele.
Imbangi ye-GPT-6 Astra kanye ne-Claude Opus 5.5
Ukwethulwa kwe-Gemini 4 Argon kunomgomo ocacile: ukuncintisana ne-OpenAI kanye ne-Anthropic . Izilinganiso ze-Google zibeka imodeli phambili kune-Claude Opus 5.5 kanye ne-Fable 5.1 ezivivinyweni ezahlukahlukene. Ku-Artificial Analysis Intelligence Index, umhloli ozimele, i-Argon ifinyelela amaphuzu angu-53, afana ne-GPT-6 Astra, imodeli ye-OpenAI esezingeni eliphezulu.
Ekuhlolweni kwe-DeepSWE, okugxile kubunjiniyela besofthiwe bangempela, i-Argon ithole amaphuzu angu-77,9% , uma kuqhathaniswa no-74,1% we-GPT-6 Astra kanye no-74,2% we-Claude Opus 5.5. I-Google iphinde yagqamisa izinga layo lokuqala ku-Vals index, elinganisa umthelela wezomnotho ezindaweni ezifana nezezimali, umthetho, kanye nentela. Futhi ku-CWE-bench, ukuhlolwa kokulungiswa kobuthakathaka, ithole amaphuzu angu-68%, okulingana nendawo yokuqala.
Enye yezinto eziphawuleka kakhulu ukwehla kokungaboni kahle . Ngokusho kwe-Artificial Analysis, izinga le-Gemini 4 Argon lingu-15% ku-AA-Omniscience, kuyilapho izinga le-GPT-6 Astra likhuphukela ku-50%. Umehluko useqinisweni lokuthi i-Argon inamathuba amaningi okuphendula ngokuthi "Angazi" uma ingaqiniseki, kunokuba iqambe impendulo. Kodwa-ke, i-Astra ifinyelela ukunemba okukhulu esivivinyweni esifanayo, ngakho-ke le datha idinga ukufakwa kumongo.
Amakhono obuchwepheshe: amathokheni ayisigidi kanye nokuzivikela kwe-cyber okuzimela
Esinye sezici ezintsha ezibaluleke kakhulu ze-Gemini 4 Argon yikhono layo lokukhiqiza amathokheni afinyelela esigidini ngempendulo eyodwa . Ukuze sikubeke kahle lokhu, ithokheni lilingana nengxenye yegama, kanti amathokheni ayisigidi angamagama angaba ngu-750.000 ngesiNgisi. Lokhu kumelela ukwanda okuphindwe kayishumi nanhlanu ngaphezu komkhawulo wangaphambilini wamathokheni angu-64.000. Ngomthamo omkhulu kangaka, imodeli ingaphatha imisebenzi eyayidinga ukuhlukaniswa ibe yizicelo eziningi, njengokuthutha i-codebase yonke.
Ukuphepha kwe-Cyber kungenye indawo lapho i-Google igxile khona emizamweni yayo. I-Argon iyakwazi ukuthola, ukuqinisekisa, nokulungisa amaphutha ezilimini zokuhlela ezingaphezu kwamashumi amabili, futhi ingahlaziya ngisho nezinhlelo zewebhu ezibukhoma ngaphandle kokudinga ukufinyelela ikhodi yazo yomthombo. Le nkampani ithi ingathola ubuthakathaka obubalulekile futhi ibulungise ngokuzimela, okufana nalokho imodeli ye-Gemini 3.8 Flash Cyber ehlose ukukwenza ukuqinisa uhlelo , yize imininingwane ethile yalokho okutholakele ingakadalulwa emphakathini ngaphezu kwalokho i-Google ekhethe ukukwabelana ngakho.
Ngaphakathi kwe-Google, izinkulungwane zabasebenzi sezivele zisebenzisa i-Argon nsuku zonke. Le nkampani ithi amaqembu ayo ayisebenzisela imisebenzi yocwaningo kanye nokuhlela, nokuthi ikhulule ama-terabyte angaphezu kuka-330 enkumbulo ezikhungweni zayo zedatha. Lezi yizibonelo ezisebenzayo inkampani ezibonisayo ukukhombisa ukuthi akuzona nje izibalo ezingabonakali, yize ulwazi lomsebenzisi lwangempela lusazobonakala.
Ukukhishwa okulinganiselwe, amanani, kanye nokungabaza ngokusebenza kwangempela
Naphezu kokuduma kwabezindaba, iqiniso liwukuthi cishe akekho ozokwazi ukuzama i-Gemini 4 Argon okwamanje . I-Google ikhethe ukuqaliswa kancane kancane, iqala ngeqembu elincane lochwepheshe bezokuphepha kwe-inthanethi abakhethwe ngohlelo lwe-Fairwind. Le modeli nayo iyahlolwa nguhulumeni wase-US, inqubo ejwayelekile ngaphambi kokuqalisa lolu hlobo lwezinhlelo. Kamuva, izotholakala kumakhasimende e-API akhokhayo asevele engababhalisile be-Google AI Ultra, yize kungekho usuku oluqondile lokukhishwa kwayo okuvamile.
Intengo yokuqala ingenye yezimpikiswano ze-Google zokuheha izinkampani. Ngesikhathi sokukhushulwa, kuzobiza u-$2 ngesigidi samathokheni ukungena kanye no-$10 ngesigidi ukuphuma (cishe u-€1,75 kanye no-€8,75). Ngemuva kwalokho, imali izophindeka kabili ibe ngu-$4 kanye no-$20. Ngokusho kwe-Artificial Analysis, ngalesi saphulelo sokuqala, izindleko zomsebenzi ngamunye zizoba ngaphansi ngo-40% kune-GPT-6 Astra.
Akuzona zonke izindaba ezinhle. I-Bloomberg ikhulume nabasebenzi be-Google abathi, yize i-Argon ikhanya ngama-benchmark, ayisebenzi kahle njengoba kulindelekile emisebenzini yokuhlela yangempela . Le nkampani iyakuphika lokhu okushiwoyo futhi igcizelela ukuthi amaqembu ayo ayisebenzisa kakhulu ngemiphumela emihle. Lokhu kungafani kubonisa iqiniso elingajabulisi: ama-benchmark okwenziwa awafani ngaso sonke isikhathi nomsebenzi wansuku zonke, lapho izindawo zokugcina izinto zinkulu khona, ukuncika kungaphelele, futhi imibhalo imbalwa.
Umongo wezepolitiki nawo ube wusizo. Ngosuku olungaphambi kwesimemezelo, i-CEO ye-Google uSundar Pichai, kanye noMongameli wase-US uDonald Trump kanye nabanye abaphathi bezobuchwepheshe, basayine isivumelwano sokuzithandela sokuzilawula. Lo mbhalo uchaza ukuhlolwa kwamabhuku kanye nokulawulwa kwangaphakathi kokuphepha kwe-inthanethi kanye nezinsongo zezinto eziphilayo. I-Google igcizelela ukuthi i-Argon inezivikelo zomngcele futhi ngeke ikhishwe kubasebenzisi kuze kube yilapho inikeza ukuphepha okuphezulu.
Ukufika kwalolu hlobo kuza ngemuva kokuba iGemini 3.5 Pro eyayithenjisiwe ingazange ibone ukukhanya kwelanga. Leyo phrojekthi yabhekana namaphutha okuqeqesha futhi yenza kahle kakhulu uma kuqhathaniswa nomncintiswano, okwaholela inkampani ekuyilahleni futhi yagxila ku-Argon, ebilokhu ithuthukiswa kusukela ngoJulayi. Ngalesi sinyathelo, i-Google ihlose ukukhombisa ukuthi isalokhu ingumncintiswano emncintiswaneni we-AI esezingeni eliphezulu, yize ukungabaza ngegebe phakathi kwezilinganiso kanye nokuhlangenwe nakho kwangempela kuqhubeka nokuhlala embonini.
