I-OpenAI yethule imodi entsha ye -Ultrafast ye-GPT-5.6 Sol yayo eyinhloko, ithembisa ukwandisa isivinini sokucubungula izikhathi eziyishumi nane. Lokhu kuthuthukiswa kudlula ukulungiswa okulula kokusebenza; kuhlose ukuqeda ukuhweba okuvamile phakathi kwesivinini nomthamo, kunikeze izimpendulo ezicishe zifane ngokushesha ngaphandle kokubeka engcupheni ubuhlakani bemodeli ethuthuke kakhulu yenkampani.
Ubuchwepheshe obusha, obutholakala njengamanje njengesiqalo eqenjini elikhethiwe lamakhasimende, busebenzisa ingqalasizinda yeCerebras, umenzi we-chip ochwepheshe ekuqondeni kwe-AI. Ngokusho kwedatha enikezwe yi-OpenAI, uhlelo lungakhiqiza amathokheni afinyelela ku-750 ngomzuzwana , okuyisibalo esikhulu ngokuphindwe kane kunezinhlobo eziningi zezentengiselwano zamanje futhi lubeke lobu buchwepheshe ezingeni elisha ngaphakathi kwemboni.
Ijubane elikhulu ngaphandle kokulahlekelwa imodeli
Kuze kube manje, ukuthola impendulo esheshayo ngokuvamile kusho ukusebenzisa imodeli encane noma ekhethekile kakhulu, isivumelwano i-OpenAI ebilokhu ifuna ukusigwema nge-Ultrafast. Esikhundleni sokudala inguqulo ehlukanisiwe, inkampani ikhethe ukusheshisa i-GPT-5.6 Sol efanayo , ishiya amakhono ayo okucabanga kanye nokukhiqiza umbhalo engashintshi. Lokhu kusho ukuthi onjiniyela nezinkampani ezisebenzisa i-API bazokwazi ukujabulela ukubambezeleka okuncane ngaphandle kokulahlekelwa ikhwalithi yemiphumela.
Ijubane lamathokheni angu-750 ngomzuzwana lisho ukuthi isigaba sempendulo sikhiqizwa ngaphansi komzuzwana, okwenza ukusebenzisana kucishe kube ngokushesha. Lesi sici sibaluleke kakhulu ezinhlelweni zokusebenza lapho yonke imizuzwana ibalwa khona, njengokuhweba nge-algorithmic, ukwesekwa kwamakhasimende okunemibuzo eminingi, noma impendulo yezehlakalo zokuphepha. I-OpenAI igcizelela ukuthi ijubane alisadingi ukuhlakanipha okudelayo , umlayezo oye wathinta abasebenzisi bokubuka kuqala.
Ingqalasizinda yeCerebras, isihluthulelo sokusebenza kahle
Imfihlo ngemuva kwe-Ultrafast isekubambisaneni kwayo ne-Cerebras, lapho iprosesa yayo ye-Wafer-Scale Engine ihlanganisa khona ama-GB angu-44 e-SRAM futhi igcina izisindo zemodeli ziseduze namayunithi e-compute. Lo mklamo unciphisa kakhulu ukunyakaza kwedatha phakathi kwememori yangaphandle neprosesa, enye yezingqinamba ezinkulu ekucabangeni ngamamodeli amakhulu. Ngokusho kwenkampani, izisindo zihlala ku-chip kuyilapho amathokheni edlula ezingqimbeni , okuvumela ukuthi iphathe amakhulu amathokheni ngomzuzwana ngaphandle kokuwohloka.
Ukuhlanganiswa neCerebras akugcini nje ngokuzuzisa i-OpenAI kodwa futhi kuphawula ingqophamlando yomenzi we-chip, obona ingqalasizinda yakhe isekela imodeli yomngcele ngaphakathi kwensizakalo yezentengiselwano. Nakuba umthamo usalinganiselwe, zombili izinkampani ziqinisekisile ukuthi zizokwandisa ukufinyelela njengoba izinsiza zitholakala. Njengamanje, i-Ultrafast inikezwa kuphela nge-API futhi ayitholakali ku-ChatGPT, umehluko i-OpenAI ewubeke kwacaca esitatimendeni sayo sabezindaba.
Izimo zokusetshenziswa ezenzelwe izindawo ezibucayi
I-OpenAI ithole izimo eziningana lapho ukubambezeleka okuphansi kwenza umehluko omkhulu. Lokhu kufaka phakathi impendulo yezehlakalo , lapho i-ejenti ingahlaziya khona amalogi, ibukeze imikhondo, futhi iphakamise izixazululo ngenkathi ukungasebenzi kusasebenza; ucwaningo lwezezimali , oludinga ukucubungula izimpawu zemakethe ngesikhathi sangempela; kanye nensizakalo yamakhasimende , lapho izingxoxo zihamba khona ngaphandle kokulibaziseka okubonakalayo. Kukhulunywa futhi nge-e-commerce, ngezincomo ezenziwe ngezifiso kanye nokuphathwa kwempahla bukhoma.
Izinkampani ezifana neJane Street, iPodium, iBasis, neRogo sezivele zihlola i-Ultrafast ezinhlelweni zokusebenza ezihlobene nokuhlela, ukuhweba, kanye nokusekelwa. Imiphumela yokuqala ikhombisa ukuthi ukunciphisa isikhathi esichithwa emsebenzini ngamunye kungaguqula izinqubo ezazisebenza ngamaqoqo ebusuku zibe yiziphindaphindo eziqhubekayo phakathi nosuku . Kodwa-ke, i-OpenAI ixwayisa ngokuthi ukusebenza kokugcina kuncike ebunzimeni bomsebenzi ngamunye, kufaka phakathi amathuluzi angaphandle kanye nezingcingo zedathabheyisi, ngakho-ke inani lamathokheni angu-750 ngomzuzwana liyiphuzu lokubhekisela kuphela lesilinganiso esiphezulu sokukhiqiza.
Ukutholakala okulinganiselwe kanye nezinyathelo ezilandelayo
I-Ultrafast okwamanje isesigabeni sokubuka kuqala esinqunyelwe, futhi i-OpenAI ayikakamemezeli usuku lokutholakala kwayo okuvamile. Le nkampani iqinisekisile ukuthi ukufinyelela kuzokhula njengoba umthamo ukhuphuka , kodwa okwamanje, iqembu elikhethiwe lamakhasimende kuphela elingakusebenzisa. Intengo yomphakathi yaleli zinga lesevisi elisha nayo ayikadalulwa, okuyisici esibalulekile kumabhizinisi ukuhlola ukuthi ukubambezeleka okuncishisiwe kuyayithethelela yini izindleko ezengeziwe.
Uma kuqhathaniswa nomncintiswano, i-OpenAI ithi i-Ultrafast idlula imodi esheshayo ka-Anthropic kaClaude, engafinyeleli isivinini esifanayo. Lo mehluko ungaba yinto enqumayo ezinhlanganweni ezithembele ezimpendulweni ezicishe zifane. Okwamanje, abathuthukisi sebevele becabanga ngezinhlelo zokusebenza ezazingasebenzi ngaphambilini ngenxa yezikhathi zokulinda, kusukela ekwakhiweni kwekhodi yesikhathi sangempela kuya emidlalweni esekelwe kumbhalo exhumanayo.
Ngamafuphi, ukufika kwe-Ultrafast kuphawula uguquko emjahweni we-AI esheshayo emakethe. Ngokubambisana kwe-Cerebras kanye nendlela enganciphisi amandla emodeli, i-OpenAI imise ngokusobala isilinganiso endaweni yokubambezeleka okuphansi. Nakuba amanani kanye nokutholakala okuvamile kusalindile, ukubuka kuqala sekuvele kukhombisa ukuthi isivinini singahambisana nobuhlakani , nokuthi imizuzwana elondolozwe emsebenzini ngamunye inenani langempela ezindaweni zebhizinisi.