Mhla ziyi-15 Kholwane 2026, i-Thinking Machines — inkampani leholwa nguMira Murati, lobekasakhulu lowengamele tobuchwepheshe e-OpenAI — yakhipha i-model yayo yekucala ye-AI, lebitwa ngekutsi yi-Inkling: i-model lenesisindvo lesivulekile lekungadawunilodwa, kusetjentiswe, futsi kuguculwe ngunoma ngubani. Imphendvulo lecondzile kumnikati webhizinisi: kubaluleka kwe-Inkling akukho ekutseni ngiyo lenemandla kakhulu — inkampani ngokwayo itsi ayisiyo lenemandla kakhulu uma kubukwa konkhe — kodvwa kukutsi yi-model lenelikhono longaba nelikhophi layo, uyisebentise kuma-server akho kuze lwati lwakho lungaphumi nanini, futsi uyilungise ngelwati lwenkampani yakho kuze isebente kangcono ngetindleko letiphansi. Yile kanye lebalulekile ebhizinisini laseSaudi lelifuna i-AI leyimfihlo lenetindleko letilawulekako.

Iyini i-Inkling, ngamafuphi?

I-Inkling isebentisa luhlelo lwe-Mixture-of-Experts: emapharamitha langemabhiliyoni langu-975 sewonkhe, kodvwa yenta kutsi kusebente kuphela lacishe abe ngemabhiliyoni langu-41 kuwo emsebentini ngamunye, loku kuyenta ibe lula futsi ishibe kuyisebentisa kunaloko lokuboniswa bubukhulu bayo. Yacecesheliswa ngema-token langematriliyoni langu-45 emibhalo, etitfombe, emsindvo nemavidiyo, futsi iwacondza ngalokwemvelo onkhe lamahlobo lamane elwati, hhayi ngetengeto letinamatseliswe ngemuva. Isekela lifasitelo lekucondza lelifika ku-1 sigidzi sema-token — loku kwenele kutincwadzi netivumelwano letiphelele esicelweni sinye — futsi tisindvo tayo letiphelele tiyatfolakala kutsi tidawunilodwe ku-Hugging Face, kantsi ingalungiswa esikhungweni se-Tinker senkampani. I-Thinking Machines iphindze ikhombise kucala inguculo lelula, i-Inkling-Small, lenemapharamitha langemabhiliyoni langu-276 sewonkhe, kulabo labafuna tindleko tekusebenta letiphansi.

"Sisindvo lesivulekile" tindzaba tangempela ebhizinisini lakho

Linyenti lema-model lahamba embili (ye-OpenAI, i-Anthropic ne-Google) avaliwe: utfumela lwati lwakho ku-server yalohlinzeka ngelusito bese ukhokha ngekusebentisa ngakunye. I-model lenesisindvo lesivulekile, ngalokwehlukile, ikunika lifayela le-model ngalokwalo, ngako-ke uyisebentisa ngekhatsi kwengcalasizinda yakho noma efwini lelingekhatsi eMbusweni. Kungani loku kubalulekile? Ngobe lwati lwakho — tivumelwano temakhasimende, emarekhodi, timfihlo tebhizinisi — lusala nawe futsi alundluleli kulomunye umuntfu, loku kwenta kube melula kutsi ulandzele uMtsetfo Wekuvikela Lwati Lwemuntfu (PDPL) lolawulwa Ngumbuso WaseSaudi Wetelwati Nebunhlakanipho Lobentiwe (SDAIA). Iphindze ikunike kutimela ngaphandle kwemhlinzeki munye: akukho kukhuphuka kwemanani ngalokungalindzelekile, futsi akukho kuvalwa kwelusito lokungamisa umsebenti wakho ngebusuku bunye.

Kuyilungisa ngelwati lwakho = kusebenta lokuphakeme ngetindleko letiphansi

Umlayeto we-Thinking Machines ucacile: yatsembela ekulungiseni esikhundleni se-model "yinye lefanele wonkhe muntfu". Ngesikhungo se-Tinker ungayicecesha kabusha i-Inkling ngelwati lwemkhakha wakho — lulwimi lwemakhasimende akho, tincwadzi takho, tinchubomgomo takho — kuze emsebentini wakho lotsite ibe nekunemba lokukhulu kune-model levamile lenkhulu, futsi ngetindleko tekusebenta letiphansi, ngobe usebentisa i-model leyenteleke sidzingo sakho esikhundleni sekubita i-model lenkhulu kakhulu emakethe kuso sonkhe sicelo lesincane. Lomcondvo — i-model lencane lelungisiwe lephumelela endzaweni yakho kune-model levamile lenkhulukati — ngiwo umgogodla welusito lolubonakalako kumanyenti etinkampani, hhayi kuncintisana kuze ube ngewekucala etinhlwini te-benchmark.

Emahlobo lamanyenti: umbhalo, titfombe nemsindvo ku-model yinye

Njengobe i-Inkling iwacondza ngalokwemvelo umbhalo, titfombe, umsindvo nevidiyo, kusetjentiswa kwayo lokubonakalako kuyandza: kuhlatiya i-invoyisi noma incwadzi leskeniwe, kubhala phansi umlayeto wemphimbo wekhasimende nekucondza injongo yawo, kubuyekeza sitfombe semkhicito, noma kukhipha lwati esivumelwaneni lesiku-PDF. Esikhundleni sekuhlanganisa emathuluzi lamanyenti lahlukene kuloluhlobo ngalunye lwalokufakwako, i-model yinye iphatsa onkhe lamahlobo — loku kwenta kube lula kwakha emasistimu phezu kwayo futsi kunciphise tindzawo lekungenteka kuyo kwehluleka.

Kodvwa "sisindvo lesivulekile" akusho kutsi akukhokhelwa ngempela

Kwetsembeka kubalulekile lapha: kudawunilowuda i-model ngaphandle kwenkhokhelo akusho kutsi kuyisebentisa akukhokhelwa. Kusebentisa i-model lenkhulu kanje kudzinga ihadiwe lenemandla (ema-GPU lambiza kakhulu) nelikhono lekusebenta kuze uyilungise, uyicaphelisise futsi uyivikele. Inkampani ngokwayo iyavuma kutsi i-Inkling ayisiyo lenemandla kakhulu uma kubukwa konkhe uma icatsaniswa nema-model lavaliwe lamasha kakhulu. Ngako-ke, ekusebentiseni ngalesinye sikhatsi, ekucaleni ngekushesha ngemcondvo, noma emisebentini levamile lelukhuni kakhulu, i-API levaliwe lesele ilungile ingahle ihlale ilula futsi ifaneleke kakhulu. Sincumo lesilungile akusiko kutsi "levulekile ihlala incono", kodvwa kutsi "yini lencono esimeni sakho".

Kufanele uyikhetse nini i-model levulekile lefana ne-Inkling?

Khetsa i-model levulekile nangabe loku kuhlangana ndzawonye: lwati lolubucayi noma lolulawulwa yimitsetfo lolungakafaneli kuphuma enhlanganweni; kusebentisa lokukhulu nalokuhlala kunjalo lokwenta kukhokhelisa ngesicelo ngasinye kubite kakhulu ngekuhamba kwesikhatsi; sidzingo sekulungisa ngalokujulile kuze kuvumelane nemkhakha wakho nelulwimi lwemakhasimende akho; kanye nelikhono lekusebenta ngekhatsi enhlanganweni noma umlingani wetobuchwepheshe kuze aphatse kufakwa kwayo. Kodvwa nawusacala noma uhlola umcondvo, cala nge-API lesele ilungile, bese udlulela etisindvweni letivulekile nangabe sidzingo sesivutsiwe, bunyenti bukhulile, futsi konga kanye nemfihlo sekukuvumela kutsi utfutse imali.

Kuzuza njani ku-Inkling nge-Origami

Ku-Origami siwaphatsa ema-model lavulekile njengethuluzi lebunjiniyela, hhayi njengesihloko lesicwazimulako: kucala sihlola kutsi i-Inkling iyafaneleka yini esimeni sakho uma icatsaniswa nalokunye lokuvaliwe; bese — nangabe ngiyo lefaneleke kakhulu — siyifaka endzaweni leyimfihlo ngekhatsi kwengcalasizinda yakho, siyicecesha kabusha ngelwati lwakho, siphindze siyihlanganise nemasistimu akho (umsiti wangekhatsi lophatsa tincwadzi takho, kwenta umsebenti ngemshini, noma kucubungula lokufakwako kwemahlobo lamanyenti), ngetilawuli tekuvikeleka netekubusa letigcina kulandzela kwakho imitsetfo. Inhloso ngumphumela lobonakalako lolinganisekako, hhayi kwemukela buchwepheshe ngenca yabo nje.

Sigcino

Kwetfulwa kwe-Inkling kucinisa indlela letsatfwa tinkampani lesitakako: ema-model lavulekile lanemandla longaba nawo, uwasebentise ngasese, futsi uwalungise ngelwati lwakho. Akusiko kukhetsa lokuncono kakhulu kuso sonkhe simo, kodvwa kuvula umnyango wangempela we-AI leyimfihlo, lenetindleko letilawulekako, nalengatsembeli kumhlinzeki munye — loku kuvumelana netinhlangano letinyenti taseSaudi letifuna kuzuza ku-AI ngaphandle kwekudzela lwati lwato.

Imitfombo

  • Thinking Machines Lab — simemetelo lesisemtsetfweni se-Inkling: https://thinkingmachines.ai/news/introducing-inkling/
  • Hugging Face — likhasi le-model i-Inkling netisindvo tayo: https://huggingface.co/blog/thinkingmachines-inkling
  • Kubikwa kwekwetfulwa (mhla ziyi-15 Kholwane 2026): TechCrunch ne-Axios. Tibalo netincazelo tinjengobe kushiwo emitfonjeni lesemtsetfweni ngelilanga lekushicilelwa.