OpenAI DevDay 2026: Izimemezelo ezingaphezu kuka-20, kusukela kumachashazi kuya ku-GPT-6.1 Sol

Lokho okuhlanganiswa yile vidiyo

• Ekungeneni okuhamba kancane kwasekuseni: i-ejenti ehlala ikhona, bukhoma esiteji
• Amachashazi: ama-ejenti ahlala evuliwe ane-khompuyutha yawo
• I-Space, Amakhasi nama-slide: idokhumenti eyodwa yabantu nama-ejenti
• I-GPT-6.1 Sol: cishe i-Astra ngesihlanu sentengo
• I-Ultrafast kanye ne-Pro 500: izikhathi eziyisithupha intengo, “kufanelekile”

Umbhalo obhaliwe ohunyushwe

Kuhunyushwe kusuka ekulandiseni kokuqala kwesiNgisi. Umsindo namazwibela atholakalayo alawulwa yi-YouTube.

Ekungeneni okuhamba kancane kwasekuseni: i-ejenti ehlala ikhona, bukhoma esiteji

0:00 Ungacabanga ukuthi i-ejenti esebenza ubusuku nemini ingabe ivukile yini ukwethulwa kwayo. I-OpenAI yachitha ihora yethula eyodwa, yabe isiyibuza umbuzo, bukhoma esiteji.

0:08 Ividiyo eyinhloko: Kusahlolwa. Ngicabanga ukuthi mhlawumbe uDotty uhamba kancane ekuseni.

0:17 I-OpenAI ithi yenza izimemezelo ezingaphezu kwamashumi amabili. Kulolu video, imibuzo emithathu. Liyini iphoyinti? Ingabe i-Sol entsha ishibhile kangakanani, ngempela? Futhi yini abathuthukisi abayitholayo ngempela ngoMsombuluko? Futhi ekugcineni, othile esiteji wacindezela inkinobho yangempela, futhi bathi yayingeyabo bonke abantu emhlabeni. Ngizokwenza lokho.

0:35 NgoLwesithathu, zingama-30 kuMandulo, futhi lokhu i-The Daily Diff. NgoLwesibili eSan Francisco, i-OpenAI yenza i-DevDay ka-2026, futhi ikhasi lesifingqo libala ukwethulwa okukhulu okungaphezu kwamashumi amabili. Iningi lazo lingena emigqeni emihlanu, futhi omunye waleyo migqa idemo osanda ukuyibuka. Qala ngamachashazi.

Amachashazi: ama-ejenti ahlala evuliwe ane-khompuyutha yawo

0:52 Iphoyinti i-ejenti esebenza ku-GPT-6 Astra, nekhanda layo efwini kanye nezithasiselo ezinhlelweni zokusebenza ezingaphezu kweziyizinkulungwane ezine. Uliyithola ku-ChatGPT, i-Slack noma i-Teams, futhi ungabiza kuyo. Ukuthumela imilayezo kuzofika maduze, ngakho i-ejenti yakho izokwazi ukukuthumelela imilayezo ngo-3 ekuseni, njengowayengumuntu wakho. I-OpenAI film yokuqalisa ibonisa ukuthi kuzwakala kanjani lokho. Umsebenzisi ulihlela kabusha ngosuku lokuqala. Ngizokubiza u-Alfred.

1:15 U-Alfred ubika ukuthi umdayisi wekhekhe lomshado ukhansele, nokuthi usevele wathola umsebenzisi obuyiselwe, okungaphezu kwalokho uhlelo lwami lwekhalenda olwake lwenzela mina. I-OpenAI iphinde ithi onjiniyela bayo banamachashazi abo alungisa inqwaba yamaphutha ngosuku. U-Holly, esiteji, wengeza ngokuthi amachashazi akhiwe ngamachashazi, okungachaza ukuhamba kancane ekuseni. Kungakapheli ihora, i-Hacker News ibinamakopi ayo amabili omthombo ovulekile, inguqulo yabathuthukisi yokukikizela kakhulu.

1:39 Iphoyinti lakho lokuqala liza namapulani e-Pro nese-Business Premium ezifanele izimakethe, futhi ukukhuluma nalo akubalwa emikhawulweni yakho yokusebenzisa. Ezinkampanini kunamachashazi akhethekile, ane-identity yawo kanye nezindawo zokusebenzela, aphathwa nge-Agent 365 yakwa-Microsoft. Izolo ngikutshele ukuthi uhulumeni wase-Dutch wakha i-Linux yawo ukuze aqhele ku-Microsoft. Namuhla ama-ejenti e-OpenAI athola umphathi wakwa-Microsoft.

I-Space, Amakhasi nama-slide: idokhumenti eyodwa yabantu nama-ejenti

2:02 Bese i-office suite. I-ChatGPT Space yikhaya elabiwe leqembu lakho, amachashazi abo kanye ne-ChatGPT, yakhelwe ku-Pages, uhlobo olusha lwedokhumenti lwabantu nama-ejenti ndawonye. Izethulo zokubambisana zilandela emavikini azayo, ngokukwazi ukukhipha ku-PowerPoint kanye ne-Google Slides, ngakho i-Notion ne-Google Docs sezithole isaziso, kudokhumenti eyakhiwe i-ejenti yakho. Manje imodeli.

I-GPT-6.1 Sol: cishe i-Astra ngesihlanu sentengo

2:22 I-GPT-6.1 Sol ibiza amadola amabili ngezithombe eziyisigidi ezingena futhi ayishumi amadola aphumayo. I-OpenAI ithi lokho kuthenga ubuhlakani obucishe bube be-Astra ngesihlanu sentengo ye-Astra. Okokufaka okugciniwe kwehle kwaba amasenti ayishumi, ingxenye yalokho i-Sol endala eyayikhokhisa, futhi omunye umsebenzisi we-Hacker News wakubiza lokho ngokuthi isimemezelo esikhulu ngempela. Iku-API, i-Codex kanye ne-ChatGPT Work, futhi ingxoxo elula ye-ChatGPT isayilindile. Ephuzwini elilodwa eledlule ngikutshele ukuthi i-Claude Opus 5.5 yanciphisa izintengo zayo, futhi i-OpenAI yaphendula nge-GPT-6 Sol ngamadola amabili.

2:55 Lowo Sol yingoweveza entsha izinsuku eziyisikhombisa ngqo, futhi ongena esikhundleni sakhe ubiza amadola amabili afanayo ncamashi. I-Hacker News yabamba umbuzo wokuthi ngabe iyona impilo yemodeli emfushane kakhulu yini. Wonke amashadi kulelo khasi yi-OpenAI ekala i-OpenAI, futhi omunye umbhalo ophansi uze uvuma inombolo ye-Fable yakwa-Anthropic ikhipha izindleko zokuwa, okwenzeka cishe kumaphesenti angamashumi amane emisebenzi. Ngakho-ke nansi ibhodi lomunye umuntu. Ku-Artificial Analysis, i-Sol ishaya amashumi amahlanu nambili kanti i-Astra amashumi amahlanu nantathu.

3:22 Izindleko ze-Sol ngomsebenzi cishe ziyingxenye yesihlanu ye-Astra, ngakho isimangalo siyaphumelela. Ibhodi elifanayo lisafaka i-Claude Opus 5.5 phezulu, kuma-58.

I-Ultrafast kanye ne-Pro 500: izikhathi eziyisithupha intengo, “kufanelekile”

3:31 Bese ijubane. I-Ultrafast isebenza i-Astra ngesivinini esifika kuma-tokens angamakhulu amathathu ngesekhondi. Ikhasi lentengo le-OpenAI libala ukuthi izikhathi eziyisithupha izinga elijwayelekile, okungamadola angamakhulu amathathu ngezithombe eziphumayo eziyisigidi. Esiteji, yonke inkulumo yalolo hlobo lwentengo yayiwumugqa owodwa. Uyangazi? Kufanele. Idemo kwakungamarokhethi amabili ajahayo, futhi elibiza kakhulu lanqoba. I-Ultrafast ku-ChatGPT idinga uhlelo olusha oluphakeme kakhulu, i-Pro 500,

3:54 okuthi i-OpenAI ithi ithola izikhathi ezingamashumi amabili nanhlanu isabelo se-Plus. Futhi uhlelo lwe-Pro olungamadola angamakhulu amabili luvulwa kabusha namuhla. Emasontweni amathathu edlule ngikutshele ukuthi i-OpenAI imise ukubhaliswa okusha kwe-Pro ngoba wonke umuntu wayefuna i-Astra. Igumbi lokulinda livaliwe, futhi kukhona igumbi elikhudlwana phezulu. Kubathuthukisi, okwenza kube yimpumelelo kakhulu yi-Decisions API.

I-Decisions API kanye ne-“AI research intern”

4:13 Unikeza imodeli encane ye-Luna uhlu olungaguquki lwezimpendulo, ikhetha eyodwa engxenyeni yesekhondi, futhi uyisebenzisa ukuhambisa izicelo, ukuhlukanisa okuqukethwe noma ukukhetha isinyathelo esilandelayo se-ejenti. Esiteji bathembisa ukuthi ividiyo yayisesikhathini sangempela, okuyilokho kanye abantu abakushoyo uma kubukeka kusheshe kakhulu. Iku-preview elinganiselwe namuhla. Bese u-Sam Altman ethatha umjikelezo wokuwina. Ngonyaka odlule i-OpenAI yabikezela umfundi ocwaningayo we-AI phakathi nonyaka,

4:35 futhi uthi manje banayo eyodwa. Umholi wocwaningo wengeze ngokuthi kusukela ngoJulayi, amamodeli abo aqedela ngaphezu kwengxenye yesithathu yemisebenzi yocwaningo lwezinsuku eziningi ngaphandle kosizo. Alikho iphepha okwamanje, ngakho-ke okwamanje umfundi ukhona ngokusho komqashi wakhe, futhi njengawo wonke umfundi, uzosolwa ngakho konke. I-Codex yathola ukwethulwa okuningi.

I-Codex: ifu, ukubuyekezwa, ukuphepha kanye “nobunzima bezwi”

4:52 Manje isebenza efwini ngezindawo ezingasetshenziswa kabusha, umbono we-Code Review uthatha ukucubungula okokuqala ezicelweni zakho zokudonsa, kanye ne-Codex Security Cloud ihlola wonke ama-GitHub amarepo ngesikhathi esihleliwe. Ulayini womyalo manje uthatha imiyalo yezwi, ngakho-ke ngokwemvelo, idemo yezwi yahamba kanje.

5:08 Ividiyo eyinhloko: Singase sibe nobunzima bezwi okwamanje, okudabukisayo. Kodwa, kulungile.

5:13 U-Romain, umethuli, uthayiphe umyalo esikhundleni salokho, futhi kamuva wanikeza inguqulo yefu umsebenzi owodwa wonke amaqembu e-backend aphupha ngawo.

5:20 Ividiyo eyinhloko: Futhi ngizothumela umyalo onjengokuthi, bhala kabusha yonke i-backend ku-Rust. [Ukuhleka.] Kungani kungenjalo?

I-Agents API, ukusebenzisa ikhompyutha kanye nama-ejenti e-OpenAI ku-AWS

5:29 Ngemuva kwalokho, u-Tibo Sottiaux, ophethe i-Codex, wabhala ukuthi amademo abukhoma ahlushwa ukukhipha zonke izibuyekezo ngesikhathi esisodwa. Ngaphansi kwakho konke kuhlala i-Agents API, uhlelo olufanayo olusebenzisa amachashazi, manje ngokusebenzisa ikhompyutha, ngakho i-ejenti yakho ingachofoza kusiphequluli ngenkathi I-OpenAI isebenzisa amaseva. I-Amazon ithola inguqulo yayo, i-Bedrock Managed Agents, esebenzisa ama-ejenti e-OpenAI ngaphakathi kwe-AWS. Futhi ezinkampanini ezinovalo, i-Private Intelligence yengeza ukungagcini idatha,

5:57 ngokubuyekezwa kokuphepha okungekho umsebenzi we-OpenAI ovumelekile ukukufunda. Manje lowo othinta isikhwama sakho.

Ngena nge-ChatGPT: uhlelo lwabasebenzisi bakho lukhokhela amathokheni

6:02 Nge-Sign in with ChatGPT, umsebenzisi ungena kuhlelo lwakho lokusebenza futhi asebenzise uhlelo lwakhe lwe-ChatGPT ngaphakathi kulo, ngakho uyeka ukukhokhela amathokheni abo. Abahlanganyeli abayishumi nesithupha bakhona, kuhlanganisa ne-Devin, i-Notion ne-Vercel, futhi umsebenzisi usetha ukuthi uhlelo ngalunye lungasebenzisa malini. Omunye umsebenzisi we-Hacker News wakubiza lokho ngokuthi bekumele kube njalo ngaso sonke isikhathi. Bese kusakazwa. Izandiso ze-Plugin zinikeza uhlelo lwakho lokusebenza ikhaya kusikhala se-ChatGPT nezindawo zalo, i-ChatGPT manje izosikisela i-plugin yakho phakathi kwezingxoxo, futhi imicimbi ye-MCP ivumela i-plugin ukuba ivuke uma kukhona okwenzekayo

6:33 enye indawo. Leso isitolo esinabasebenzisi abayizigidi eziyizinkulungwane eziyodwa nengxenye ngesonto, futhi umnikazi oyedwa. Futhi ukukhombisa i-API eshayela i-hardware, u-Romain waletha idada lerobhothi elibizwa Lavender.

6:44 Ividiyo eyinhloko: Hey, Lavender, ungakwazi yini ukukhala uma ungizwa? [Ukukhala. Ukuhleka.]

6:49 Manje, leyo nkinobho.

Inkinobho yokusetha kabusha, yawo wonke umuntu emhlabeni

6:51 Esiteji, i-OpenAI yahlekisa ngokuthi isenze ukusetha kabusha okuningi kangangokuthi kukhona umkhankaso woku ukuyiqamba kabusha ibe inkampani yokusetha kabusha. Bese u-Tibo, okusetha kabusha kwakhe imikhawulo yokusebenzisa kuwumdlalo phakathi kwabasebenzisi be-Codex, wacindezela inkinobho yangempela ngaphansi kwekhava ephindekayo. Umbuzo esiteji kwaba, ingabe lokhu kusebenza kanjalo nje? Bathi kwakungokwabantu bonke emhlabeni, okuyindlela isethulo esiphela ngayo ngesixuku sabathuthukisi abahlaziya ikhasi labo lokusebenzisa. Ukubuyekezwa kwahlukaniswa.

7:15 Umsebenzisi we-Hacker News wakubiza lokho ngokuthi ihihliza, futhi impendulo eyodwa ku-X yakubiza ukusetha kabusha kube yikho kuphela okuhle kosuku. Bese u-Sam Altman eva. Wathi uthola ukuqhathaniswa kwe-industrial revolution kungakhangi, nokuthi i-AI eyenziwe kahle ingaba kakhulu njenge-Renaissance entsha. I-Renaissance, enomkhawulo wokusebenzisa kanye nenkinobho yokusetha kabusha. Uma ungathanda ukufunda lokhu kunokungizwa ngikusho, i-diff ifika ebhokisini lakho njalo ekuseni, mahhala ku-the daily diff dot dev, isixhumanisi ngezansi.

7:40 Ngakho, isinqumo sanamuhla nge-DevDay.

Isinqumo: NEEDS REVIEW + uhlu lwakho lwangoMsombuluko

7:43 NEEDS REVIEW. Ngingasishaya kanjalo ngoba izintengo ziyahambisana kwenye ibhodi yokushaya, ngenkathi umfundi, amachashazi nama-demos kuseyi-OpenAI ekala i-OpenAI. Bhalisa, shaya insimbi, futhi ungitshele kumazwana uma uzo uyisishaye ngokuhlukile. Uhlu lwakho lwangoMsombuluko. Hambisa imisebenzi ye-batch ne-ejenti ku-Sol futhi uncike ku-cache eshibhile, zama i-Decisions API ukuze uhambise, futhi ubheke i-Sign in with ChatGPT ngaphambi kokuba umqhudelani wakho enze. Futhi yilokho okuhlukile kwanamuhla.

8:08 Ngingu-Niko ovela ku-Axrisi. Hlanganisa ngokuzibophezela.

Imithombo

1. DevDay 2026 recapopenai.com
2. Keynote VODwww.youtube.com
3. Introducing GPT-6.1 Solopenai.com
4. Introducing dotsopenai.com
5. API pricingdevelopers.openai.com
6. Artificial Analysisartificialanalysis.ai
7. Hacker News (Sol)news.ycombinator.com
8. Hacker News (recap)news.ycombinator.com
9. Tibo Sottiaux on the demosx.com
10. Sign in with ChatGPT help pagehelp.openai.com