GPT-6 Luna Uyasebenza

I-GPT‑6 Luna iyimodeli elula eyethulwe yi-OpenAI, futhi iyinguqulo ephuma ku-GPT‑6 Astra kanye ne-GPT‑6 Sol, egxile emisebenzini evamile eyenziwa ngezinga elikhulu. Intengo ye-API yemodeli yehliselwe ku-$0.1 kokokufaka kanye no-$0.5 kokokuphumayo ngemathokheni ayisigidi, okushibhe ngo-50% kunesizukulwane esedlule, futhi isingene emkhakheni we-DeepSeek. Ngokusebenza, ithole u-66.6% esivivinyweni sokuhlela i-DeepSWE, isondela ezingeni eliphakathi le-Claude Opus 5, kuyilapho izindleko zayo zingaba ngu-7% kuphela kwezayo. Ilungele ukuthuthukiswa okulula, ukucubungula ngenqwaba kanye nezimo zokusebenza kwama-Agent amaningi ngesikhathi esisodwa.

Ukuhlela okulula: I-DeepSWE v1.1 ithole u-66.6%, isondela ezingeni eliphakathi le-Claude Opus 5, kodwa izindleko zomsebenzi ziphansi ngo-93%.
Ukusebenza kwamakhompyutha: Esivivinyweni se-OSWorld 2.0 yeqe i-GPT-5.6 Sol (medium), ngezindleko eziyishumi kuphela zayo.
Ukugeleza komsebenzi kochwepheshe: Ngamandla aphezulu okucabanga, amaphuzu e-AutomationBench aphakeme ngamaphesenti angu-5.4 kun e-GPT-5.6 Luna, kuyilapho izindleko zomsebenzi zehle ngo-58%.
Ukunemba kwamaqiniso: Ngamandla aphezulu okucabanga, ifinyelela ezingeni lokwethembeka kwamaqiniso le-GPT-5.6 Sol ngezindleko ezicishe zibe yiphesenti elilodwa kuphela.
Ukuthuthukisa isitayela sokubambisana: Izimpendulo zicace futhi zibe mfushane, zisebenzise amagama obuchwepheshe ambalwa nemininingwane engadingekile emincane, futhi zichaze ngokusobala ukuthi yikuphi okuhloliwe.

Ukwehliselwa kwamandla e-Astra: Isebenzisa izindlela zokuqeqesha ezisondelene nalezo ze-GPT‑6 Astra, icindezela amakhono emodeli ehamba phambili ekucabangeni, emithonjeni eminingi kanye nasekuvumelaniseni abe yinguqulo elula esebenza ngokushesha, ukuze kuzuzwe “ukwandisa amandla ngokwezinga” kunokulandela kuphela ukusebenza okuphezulu kakhulu.
Ukuthuthukiswa kwengqalasizinda yokucabanga: Ngokuthuthukisa indlela yokugcina idatha yesikhashana kanye nokusebenza kahle kokucabanga, yehlisa izindleko zesevisi, bese idlulisa lokho kongiwa kubasebenzisi ngokwehlisa intengo ngo-50%. Yilokhu okuyisisekelo kwentengo ephansi, hhayi ukunciphisa amakhono emodeli kuphela.
Ukuthuthukiswa kwe-Prompt Caching: Uhlelo lunezinga eliphezulu lokuthola okufakwe kunqolobane ngokuzenzakalelayo, kanti ukufundwa kokokufaka okunqolobane kuthola isaphulelo esingu-90%; lusekela ukulungiswa kwamandla okucabanga kanye nokuvulwa nokuvalwa kwamathuluzi phakathi kwengxoxo ngaphandle kokuphazamisa ukusetshenziswa kabusha kwenqolobane, futhi lufaka izindawo zokunqamula ezicacile ukuze abathuthukisi balawule ngokunembile imingcele yesiqalo esinqolobane. Lokhu kubaluleke kakhulu kuma-Agent anomongo omude.
Amazinga ahlukene okucabanga: Lunikeza izilungiselelo eziningi zomzamo, eziphansi, eziphakathi, eziphezulu nezisezingeni eliphakeme kakhulu, ukuze abasebenzisi balungise kalula amandla asetshenziswayo kuye ngobunzima bomsebenzi. Amandla aphezulu okucabanga angasondela ezingeni le-Sol yesizukulwane esedlule, kanti amandla aphansi anciphisa izindleko nakakhulu, okwenza kube nokulinganisela okuguquguqukayo phakathi kwezindleko namandla.

Qinisekisa ukuthi ungumsebenzisi we-Plus, Pro, Business, Enterprise noma Edu (abasebenzisi be-Free/Go bangasebenzisa i-Luna kuphela ezinhlelweni zedeskithophu).
Vula i-ChatGPT Work noma i-Codex, bese uthola i-GPT-6 Luna esikhethweni samamodeli.
Faka umsebenzi ngokuqondile ukuze uqale. I-Luna ilungele imisebenzi evamile, elula neyenzeka kaningi.
Ngena kunkundla yabathuthukisi be-OpenAI ukuze uthole i-API Key.
Esicelweni, setha ipharamitha yemodeli ibe ngu-gpt-6-luna, bese ukhokha u-$0.1 kokokufaka kanye no-$0.5 kokokuphumayo kumathokheni ayisigidi.
Ungasebenzisa i-Prompt Caching Dashboard yekhonsoli ukuqapha izinga lokutholwa kwenqolobane, bese usebenzisa izindawo zokunqamula ezicacile ukuthuthukisa ukusetshenziswa kabusha komongo omude nokunciphisa izindleko.

Ukusebenza okudlula isigaba sayo: Esivivinyweni sokuhlela i-DeepSWE ithole u-66.6%, isondela ezingeni eliphakathi le-Claude Opus 5, kodwa izindleko zomsebenzi ziphansi ngo-93%.
Ukungena endaweni yamanani ye-DeepSeek: Intengo yezicelo ezivamile isingene ebangeni le-DeepSeek V4.1 Flash futhi ishibhile nakakhulu, ngaphandle kokuguquguquka kwamanani phakathi kwezikhathi ezimatasa nezingamatasa.
Ukunemba kwamaqiniso okuthuthukisiwe: Ngamandla aphezulu okucabanga, ifinyelela ezingeni lokunemba kwamaqiniso le-GPT-5.6 Sol ngezindleko ezicishe zibe yiphesenti elilodwa kuphela.
Ukonga okukhulu ngenqolobane: I-Prompt Caching ethuthukisiwe inezinga eliphezulu lokutholwa, kanti ukufundwa kokokufaka okunqolobane kuthola isaphulelo esingu-90%, okunciphisa nakakhulu izindleko kuma-Agent anomongo omude.
Ilungele ukusebenza ngezinga elikhulu ngesikhathi esisodwa: Ukwakheka kwayo okulula kunikeza umthamo omkhulu wokucubungula, okulungele imisebenzi evamile efana nokucubungula ngenqwaba, ukuthuthukiswa okulula kanye nokusebenza kwama-Agent amaningi ngesikhathi esisodwa.

Iwebhusayithi esemthethweni yephrojekthi: https://openai.com/index/introducing-gpt-6-sol-and-luna/

Ukuhlela okulula: Yenza ukuhlolwa kwekhodi, izinguquko zenqwaba kanye nokuthuthukiswa kwemisebenzi elula, inikeze amandla okuhlela asondela ezingeni eliphakathi le-Claude Opus 5 ngezindleko ezicishe zibe ngu-7% kuphela zalo.
Imisebenzi yama-Agent amaningi ngesikhathi esisodwa: Isekela ukugeleza komsebenzi okude okwenziwa ama-Agent amaningi ngokuhambisana ngezinga elikhulu, futhi intengo ephansi yenza inzuzo yezindleko ikhule ngokuphindaphindiwe njengoba inani lezicelo likhula.
Umsebenzi wasehhovisi wansuku zonke ovamile: Iphatha ukubhalwa kwama-imeyili, ukufingqa amadokhumenti, ukuhlela idatha kanye neminye imisebenzi ephindaphindwayo. Intengo ephansi engaguquki ngaphandle kokuguquguquka kwezikhathi zokuqina yenza isabelomali silawuleke ngokuphelele.
Ukusebenza kwamakhompyutha okuzenzakalelayo: Yenza ukuphequlula iwebhu, ukugcwalisa amafomu kanye neminye imisebenzi efana ne-RPA, ithole amaphuzu aphezulu okusebenza ngezindleko eziyishumi kuphela ze-GPT-5.6 Sol.
Ukusatshalaliswa kwezinhlelo ze-AI ngezinga elikhulu: Njengemodeli engemuva yama-chatbot nemikhiqizo yokukhiqiza okuqukethwe, intengo yayo ephansi kakhulu ye-API yenza izindleko ezengeziwe zezicelo ezivamile zisondele kuqanda.