Izolo (Agasti 6), i-Neon ishicilele umbhalo webhulogi, imemezela ukubambisana ne-Castform futhi yadalula uhlelo lokuqeqesha oluqondiswe kumamodeli okusesha.
Qaphela: I-Neon iyisizindalwazi se-PostgreSQL esingenaseva (Serverless), esiphethwe ngokuphelele. Ihlukanisa amandla okucubungula nokugcinwa kwedatha, isekela ukunwetshwa okuzenzakalelayo, ukunciphisa umthamo uze ufike kuziro lapho kudingeka, kanye nokuhlukaniswa kwesizindalwazi okufana ne-Git. Ifanele kakhulu izinhlelo zokusebenza zesimanje ezisebenzisa i-cloud-native kanye nama-agent e-AI.
I-Castform yona ihlinzeka ikakhulukazi ngama-agent ezwi e-AI kanye nenkundla yokuqeqesha amamodeli eqondiswe emabhizinisini nakubathuthukisi. Ngokusebenzisa ukuxhumana kwezwi okufana kakhulu nokwangempela kanye nemizila yokusebenza ezenzakalelayo, isiza amabhizinisi ukuphatha izingcingo zansuku zonke, ukulandelela ukuthengisa kanye nokweseka amakhasimende.
Lo mbhalo webhulogi uveza ukuthi i-agent ye-AI enhle idinga amakhono angu-2 alandelayo: okokuqala, ukuthola idatha efanele yomongo ngamathuluzi; okwesibili, ukunquma ukuthi iyiphi imodeli okufanele isetshenziswe kulokho kusesha.
Ngakho-ke ithimba le-Neon libambisene ne-Castform lasebenzisa ukufunda okuqinisa ukuze liqeqeshe ngemva kokuqeqeshwa imodeli yomthombo ovulekile enepharamitha engu-4B. Emisebenzini yokusesha, ukunemba kwale modeli kuyalingana noma kuphakeme kune-GPT-5.6 Sol ye-OpenAI, kanti izindleko zayo zilinganiselwa ku-1/100 kuphela.
Ithimba le-Neon liveze ukuthi endleleni yokusesha amadokhumenti, indlela ejwayelekile njengamanje ukusesha okushumekiwe (embedded search), okuvamise ukuguqula amadokhumenti abe yidatha yezinombolo bese kuseshwa okuqukethwe okufanayo.
Njengoba ama-agent e-AI esanda ukusabalala, inqubo yokusesha isiqala ukuguqukela ekusesheni okuqhutshwa yi-agent. Kulezi zinhlelo, imodeli ye-AI ihlukanisa umbuzo omkhulu ube yimibuzo emincane eminingi, izinqumele yona imibuzo yokusesha, ibuke imiphumela yokusesha, bese iqala olunye usesho ngokusekelwe emiphumeleni kuze kube ithola ulwazi oludingekayo.
I-Neon ithi ukusesha okuqhutshwa yi-agent kulungele kakhulu ukuphatha imibuzo eyinkimbinkimbi, kodwa usesho ngalunye ludinga ukubiza imodeli esebenza kahle kakhulu, ngaleyo ndlela kwandise isikhathi nezindleko. Ngokwesilinganiso sesicelo esivamile esinikezwe yi-Neon, ukusebenzisa i-GPT-5.6 Sol ekusesheni okukodwa kuthatha ngaphezu kwemizuzwana engu-10, ngezindleko ezingaba ngu-$0.03.
I-Castform ikhethe ukuqeqesha imodeli yomthombo ovulekile encane uma kuqhathaniswa, ukuze ifunde ukuthola ulwazi oludingekayo kudatha ethile. I-Neon ibhekelela indawo yokugcina amadokhumenti kanye nomsebenzi wokusesha, kanti i-Castform ibhekelela ukuqeqesha imodeli ukuthi inqume “okufanele iseshwe”.
Ngesikhathi sokuqeqesha, imodeli izama ukuqedela umsebenzi, uhlelo lunikeza imiphumela amaphuzu, bese lufaka lawo maphuzu ezivivinyweni ezilandelayo. Izindlela zokuhlola zihlanganisa: ukuthi idokhumenti efanele itholakele yini, ukuthi izigaba ezifanele zicashuniwe yini, nokuthi impendulo yokugcina ilungile yini.
Ekuqinisekiseni kwe-Neon, imodeli eqeqeshwe ngemva kokuqeqeshwa yi-Castform ithole isilinganiso samaphuzu angu-1.447, yadlula u-1.369 we-GPT-5.6 Sol, yaphinde yadlula u-1.377 we-GPT-5.4 ebisetshenziswa njengesiqhathaniso, yabeka irekhodi eliphakeme kunawo wonke kulokhu kuqinisekiswa.
Mayelana nezindleko, izindleko zokucubungula okukodwa kwemodeli eqeqeshwe ngemva kokuqeqeshwa zingu-0.000929 USD (cishe 0.0063 yuan yase-China ngokwezinga lokushintshisana lamanje); izindleko zokucubungula okukodwa kwe-GPT-5.6 Sol zingu-0.087338 USD (cishe 0.59 yuan yase-China ngokwezinga lokushintshisana lamanje).