Waxaan si helitaan guud ah u daahfuraynaa qoyska moodallada GPT‑5.6 ka dib hordhacayagii xaddidnaa: hormuudkayaga cusub, Sol, oo ay weheliso Terra, nooc dheellitiran oo loogu talagalay shaqada maalinlaha ah, iyo Luna, noocayaga ugu kharash-oolka badan.
GPT‑5.6 Sol wuxuu dejinayaa heer cusub oo ku saabsan garaad iyo hufnaan labadaba, isagoo natiijooyin heerkii ugu sarreeyay ka gaaraya barnaamij-qorista, shaqada aqoonta, amniga internetka, iyo sayniska, isla markaana ka waxqabad sarreeya noocyadii hore iyo moodallada ugu casriyeysan ee tartamaya isagoo adeegsanaya token-no yar iyo kharash la qiyaasay oo hooseeya. Natiijadu waa waxqabad ka xoog badan doolar kasta: shaqo guul badan oo lagu gaaro isla kharashkii, ama natiijooyin la mid ah oo lagu helo kharash guud oo ka hooseeya. Waxaan sidoo kale soo bandhigaynaa hab cusub oo lagu dardargeliyo shaqada ugu culus: ultra waa dejintayada awoodda ugu sarreysa, iyadoo isku dubbaridda wakiillo badan oo ka shaqeeya hawlaha isbarbar socda si hawlaha adag loogu dhammeeyo si dhakhso leh. Adeegsiga kombiyuutarka oo ka xooggan iyo garashada naqshadeynta ayaa GPT‑5.6 ka dhigaya Sol waa la-shaqeeyahayagii ugu hufnaa ilaa hadda, isagoo ka caawinaya inuu baaro, turxaan-bixiyo, oo keeno natiijooyin diyaar u ah isticmaalka.
Waxaan GPT‑5.6 u tababarnay in token kasta laga helo shaqo waxtar badan. On Imtixaankii Ugu Dambeeyay ee Wakiillada(ku furmaa daaqad cusub), oo ah qiimayn ku saabsan habraacyada shaqo ee xirfadeed ee muddada dheer socda kana kooban 55 goobood, GPT‑5.6 Sol wuxuu dejinayaa heer cusub oo sare oo ah 53.6, isagoo Claude Fable 5 (caqliyeynta la-qabsiga) kaga sarreeya 13.1 dhibcood. Xitaa heerka caqliyeynta dhexdhexaadka ah, waxay Fable 5 kaga sarreysaa 11.4 dhibcood, iyadoo ku kacaysa qiyaastii rubuc ka mid ah kharashka la qiyaasay. Hufnaantaas waxay sidoo kale ku fidsan tahay moodallada yaryar, kuwaas oo muhiim u ah in garaadka laga dhigo mid ka badan oo la awoodi karo: GPT‑5.6 Terra iyo GPT‑5.6 Luna waxay Fable 5 ka waxqabad fiican yihiin iyagoo ku kacaya qiyaastii hal lixaad (1/16) kharashka. Marka lagu eego Tusmada Sirdoonka Macmalka ee Falanqaynta(ku furmaa daaqad cusub), oo ah cabbir ballaaran oo sirdoonka ah kana kooban shaqada agentic, kood-qorista, caqliyeynta sayniska, iyo awoodaha guud, GPT‑5.6 Sol oo leh caqliyeynta ugu badan wuxuu hal dhibic u jiraa Fable 5, isagoo hawlaha ku dhammaynaya waqti 61% ka yar iyo qiyaastii kala bar kharashka la qiyaasay.
Imtixaankii Ugu Dambeeyay ee Wakiillada(ku furmaa daaqad cusub): Socod-hawleedyo wakiil-ku-salaysan oo muddada dheer ah oo ka jira qaybaha xirfadeed.
GPT‑5.6 waxa lagu daahfurayaa tallaabooyinkeennii ilaalineed ee ugu adkaa ilaa maanta, kuwaas oo loogu talagalay inay u adkaystaan si xun u adeegsi qorshaysan oo la qabsanaya, iyada oo aan si ballaaran loo xaddidin shaqada sharciga ah. Ka hor helitaanka guud, waxaan moodallada iyo tallaabooyinka ilaalinta gelinnay muddadii qiimeynta ugu ballaarneyd ilaa hadda, annagoo isku darnay qiimeynta awoodda nidaamka ee ay dadku sameeyaan iyo tijaabinta otomaatigga ah ee baaxadda weyn. Intii lagu jiray muddadii tijaabada hordhaca ah, waxaan si dhow ula shaqaynay hay’ado khubaro ah iyo la-hawlgalayaal lagu kalsoon yahay si aan si adag u tijaabino difaacyada una xoojino tallaabooyinka ilaalinta ka hor daahfurka ballaaran. Nidaamka ka dhasha waxa uu ilaalinnada lagu tababaray moodal ku lakabeeyaa hubinno waqtiga-dhabta ah, kormeer, iyo helitaan lagu habeeyey kalsoonida iyo khatarta.
GPT‑5.6 Sol waa moodalkeena koodh-qorista ugu fiican ilaa hadda. Ku saabsan Tusmada Wakiilka Falanqaynta Farsamada ee Artificial Analysis Coding Agent, GPT‑5.6 Sol oo leh caqliyeynta ugu sarreysa ayaa dejisay heer cusub oo ugu sarreeya oo ah 80, 2.8 dhibcood ka sarreeya Fable 5, iyadoo adeegsanaysa in ka yar kala bar token-nada wax-soo-saarka, qaadata in ka yar kala bar waqtiga, kharashkuna yahay qiyaastii saddex-meelood meel ka yar. Faa’iidadaas waxay ku fidsan tahay guud ahaan qoyska: Terra wax yar ayay ka sarreysaa Fable 5, halka Luna ay ka fiican tahay Opus 4.8; mid kastaana sidaas ayuu ku qabtaa qiyaastii saddex meelood meel wakhtiga, iyadoo la isticmaalayo ku dhowaad kala bar tirada token-nada wax-soo-saarka, iyo qiyaastii afar meelood meel kharashka la qiyaasay. Waxay sidoo kale dejinaysaa natiijooyin cusub oo heerka ugu sarreeya ah Terminal‑Bench 2.1 iyo DeepSWE, kuwaas oo tijaabiya socod-hawleedyo khadka amarada oo adag iyo injineernimo hiigsi-dheer oo ku jirta kayd-koodh dhab ah.
Tusmada Wakiilka Koodhaynta ee Artificial Analysis: tusme madaxbannaan oo ku saabsan waxqabadka wakiillada koodhaynta ee dhinacyada hirgelinta, adeegsiga terminalka, iyo saldhigyada koodhka dhabta ah.
GPT‑5.6 wuxuu qori karaa oo socodsiin karaa barnaamijyo fudud oo isku dubbarida qalabka, farsameeya natiijooyinka ku-meel-gaarka ah, la socda horumarka, isla markaana doorta tallaabada xigta marka shaqadu socoto. Tani waxay u suurto-gelinaysaa hawlaha ku tiirsan qalab badan inay horumar ku sameeyaan iyagoo adeegsanaya token-yada ka yar, wareegyo yar oo moodal ah, iyo hagis yar. Halkii horumariyayaasha looga baahnaan lahaa inay qoraal ku qoraan tallaabo kasta ama jawaab kasta oo qalab dib loogu mariyo moodal, Wicitaanka Qalabka ee Barnaamijaysan(ku furmaa daaqad cusub) ee Responses API wuxuu shaandhayn karaa xaddi badan oo xog dhexdhexaad ah, wuxuu hayn karaa oo keliya waxa muhiimka ah, wuxuuna la qabsan karaa hab-socodkiisa shaqo inta hawshu socoto.
Dhibaatooyinka u baahan gelin waqti iyo awood xisaabeed oo badan, GPT‑5.6 waxa uu ka gudbi karaa dejintan caadiga ah ee waxtarka leh. max wuxuu GPT‑5.6 siiyaa waqti ka badan kan xhigh si uu u fikiro, u sahamiyo beddello, u fuliyo hubinno, una dib u habeeyo habka uu u wajahayo. ultra wuxuu sii tallaabsadaa isagoo si caadi ah u isugu dubbaridaya afar wakiil oo si isbarbar socda u shaqeeya, taas oo isticmaalka token-nada oo ka badan ku beddelaysa natiijooyin xooggan iyo waqti ka dhakhso badan oo natiijo lagu helo marka la qabanayo hawlo adag. Jaantusyada hoose waxay isbarbar dhigayaan qaabka afar-wakiil ee caadiga ah ee ultra iyo halbeegga tixraaca ee hal-wakiil guud ahaan BrowseComp, SEC-Bench Pro, iyo Terminal-Bench 2.1; BrowseComp iyo SEC-Bench Pro sidoo kale waxay muujinayaan qaabeynta 16-wakiil. Dhammaan saddexda qiimayn, ku darista wakiillo isbarbar socda waxay ugu casriyeysan dhibcaha iyo daahitaanka ku riixdaa kor iyo bidix, iyadoo gaadhaysa natiijooyin xooggan waqti gaaban gudaheed. API-ga dhexdiisa, horumariyayaashu waxay dhisi karaan waayo-aragnimo la mid ah Ultra iyagoo adeegsanaya beta-da wakiillo badan ee Responses API.
GPT‑5.6 waxay keentaa isbeddel weyn oo ku yimaadda garashada naqshadaynta. Iyada oo keliya la adeegsanayo tilmaamo guud oo heer sare ah, GPT‑5.6 wuxuu abuuraa is-dhexgalyo bilicsan, u fudud isticmaalka, oo shaqaynaya. Awoodihiisa ka xooggan ee isticmaalka kombiyuutarka waxay u saamaxayaan inuu baaro oo hagaajiyo natiijada la soo bandhigay—ma aha oo keliya inuu abuuro koodhka ama nuxurka hoose—sidaas darteed wuxuu qaban karaa arrimaha muuqaalka iyo kuwa shaqeynta, wuxuuna ku dari karaa taabashooyinka ugu dambeeya ka hor inta uusan shaqada dib u celin.
Weydiin: Ma ii hirgelin kartaa ciyaar shiraacasho oo 3D ah? Wax kasta oo u baahan sawirro bitmap ah, textures (dusha sare ee walxaha), ama sprites (sawirro loo isticmaalo ciyaaraha), ama haddii ay kaa caawinayso inaad haysato tusaale/mockup muuqaal ah markaad samaynayso moodallo 3D ah, xor ayaad u tahay inaad isticmaasho Imagegen.
Awoodda dhanka naqshadaynta (frontend) ee GPT‑5.6 waxay sidoo kale u beddeshaa codsiyada ku qoran luuqadda caadiga ah mid sharraxaad iyo muuqaalo is-dhexgal ah oo aad nadiif u ah, kuwaas oo ka dhex shaqeeya ChatGPT Work.
Weydiin: Samee spirograph is-dhexgal ah si aad u sharaxdo sida uu u shaqeeyo.
GPT‑5.6 wuxuu bixiyaa natiijooyin aad u wanaagsan marka loo dhiibo hawlo u baahan xirfad wanaagsan. Wuxuu macluumaad isku qasan ka soo qaataa dukumentiyadaada iyo ababka shaqo ee sida maalinlaha ah loo isticmaalo sida Slack, Notion, Microsoft 365, iyo Google Drive, ka dibna wuxuu u beddelayaa wax-soosaarro heer khabiir ah oo la wadaagi karo.
Awoodda GPT‑5.6 ee shaqada aqoonta ayaa ka muuqata qiimayno kala duwan oo daboolaya falanqaynta xirfadeed ee muddada fog, daalacashada, adeegsiga aaladaha, iyo adeegsiga kombiyuutarka. GPT‑5.6 Sol waxay dejisaa natiijooyin cusub oo heerka ugu sarreeya ah BrowseComp oo ah 92.2% iyo OSWorld 2.0 oo ah 62.6%; OSWorld, waxay ka sarreysaa Opus 4.8 iyadoo adeegsanaysa 85% ka yar token-nada wax-soo-saarka. Halkan, faa’idooyinka waxqabadka halkii doolar waxay ku fidsan yihiin dhammaan qoyska GPT‑5.6. Luna waxay ku dhowaad la mid tahay GPT‑5.5 waxqabadka ugu sarreeya iyadoo kharashka la qiyaasay uu ka yar yahay kala bar, halka Terra ay ka sarreyso iyada oo kharashkeedu uu aad u hooseeyo.
BrowseComp: GPT‑5.6 Sol wuxuu gaaraa heer cusub oo ugu sarreeya ee BrowseComp, kaas oo ka kooban hawlo daalacasho wakiilnimo leh.
GPT‑5.6 Sol waxa uu wanaajiyaa tayada bandhigyada, dukumentiyada, iyo xaashiyaha xisaabeed, isaga oo soo saaraya natiijooyin aad u hufan isla markaana sax ah. Waxay abuuri kartaa bandhigyo si buuxda wax looga beddeli karo, laga bilaabo eber, iyadoo weydiin iyo agabka aasaasiga ah u beddelaysa sheeko muuqaal ah oo isku xiran, leh habaynno xooggan, kala-sarrayn, iyo naqshad.
Horumarku si gaar ah ayuu u muuqdaa marka la raacayo qaab-dhismeedyada iyo xirmooyinka slaaydhyada tixraaca. GPT‑5.6 wuxuu ka garan karaa nidaamka naqshadaynta ee xirmada bandhigga—qaab-dhismeedyada, qaabka qoraalka, kala-fogaanshaha, midabada, iyo qaababka nuxurka ee soo noqnoqda, oo ay ku jiraan xeerarka ku dhex jira Slide Master-ka—wuxuuna si joogto ah ugu dabaqi karaa habraacyadaas waxyaabaha cusub. Tusaalahan, marka la waydiiyo in la cusboonaysiiyo tirooyinka iyadoo lagu salaynayo fayl tixraac ah, soo-saarka GPT‑5.5 waxaa ka maqan qaybo muhiim ah oo ka mid ah slide-ka master-ka, halka GPT‑5.6 uu si ka sax ah u raacayo qaab-dhismeedka tixraaca.
Faylka tixraaca
Wax-soo-saarka GPT‑5.5
GPT‑5.5 waxaa ka maqan qaybaha muhiimka ah ee slide-ka master-ka
Natiijada GPT‑5.6
GPT‑5.6 waxay kale oo abuurtaa dukumeentiyo iyo xaashiyo xisaabeed muuqaal ahaan ka sii habaysan sidii hore. Waxay si ka sax ah u raacdaa qaabab tixraac oo adag, taas oo muhiim u ah hawlaha shaqada aqooneed ee dib loo celin karo. Waxay si saxnaan ku dheehan u maareysaa moodallada isla'egyada iyo kuwa maaliyadeed, waxayna si ka wanaagsan sidii hore uga faa'iidaysataa hab-qoraalka, kala-fogaanshaha, kala-sarraynta, iyo qaabeynta bogga ama xaashida shaqada.
Macaamiishii ugu horreeyay ee tijaabinaysay GPT‑5.6 waxay arkeen horumar ku yimid natiijooyinka shaqada aqoonta ee qaybaha kala duwan.
GPT‑5.6 waa moodalkeena amniga saybarka ee ugu xooggan ilaa hadda, isagoo gaaraya waxqabadka ugu casriyeysan adoo adeegsanaya token-no aad u yar. Marka laga hadlayo ExploitBench2, kaas oo cabbira horumarka laga bilaabo gaarista koodh nugul ilaa fulinta koodh aan xadidnayn, waxay gaartaa 73.5% marka loo eego tan GPT‑5.5 47.9% marka la joogo miisaaniyad token wax-soo-saarka oo la barbardhigi karo. Marka laga hadlayo ExploitGym3, oo wakiillada ka dalbada inay nuglaanshooyinka dhabta ah ee adduunka u beddelaan exploit-yo shaqaynaya, waxay ku dhowaad labanlaabtaa tii GPT‑5.5 heerka gudbidda ugu sarreeya, laga bilaabo 15.1% ilaa 24.9% iyadoo ku hoos jirta xadka 2-saacadood; marka lix saacadood la adeegsado, wuxuu gaarayaa 33.7%. Marka laga hadlayo SEC-Bench Pro, oo tijaabiya soo saarista tusaalaha fikradda ee software adag, waxay gaartaa 71.2% marka loo eego tan GPT‑5.5 45.8% oo leh daahitaan la hagaajiyay.
GPT‑5.6 wuxuu taageeraa hawlo difaac oo muhiim ah sida dib-u-eegista amniga ee koodhka, rakibidda balastarrada, qaabaynta hanjabaadaha, iyo hawlaha kooxda buluugga. Shakhsiyaadka iyo ururrada u qalma ee ku jira barnaamijka Helitaanka la Aamini karo ee Cyber-ka ee OpenAI Daybreak waxay heli karaan awooddeeda difaac oo dheeraad ah iyada oo loo marayo ilaalinno ka sii sax ah oo loogu talagalay shaqo la xaqiijiyay oo ka dhacaysa deegaano la oggolyahay, oo ay ku jiraan kala hormarinta iyo xaqiijinta nuglaanta, falanqaynta malware-ka, injineerinka ogaanshaha, iyo xaqiijinta patch-ka.
Shakhsiyaadku waxay xaqiijin karaan aqoonsigooda oo codsan karaan helitaan la aamini karo(ku furmaa daaqad cusub), hay'aduhuna waxay kooxahooda u codsan karaan. Xubnaha shakhsiga ah waa inay awood siiyaan Amniga Koontada ee Horumarsan(ku furmaa daaqad cusub) iyagoo adeegsanaya furayaasha gelitaanka ee ay taageeraan qalabku ugu dambayn 1-da Sebtembar si ay u sii haystaan helitaanka moodalladeena ugu casriyeysan ee ugu kartida badan dhanka saybarka; kuwa aan sidaas samayn waxay ku laaban doonaan helitaanka caadiga ah. Isticmaalayaasha aan hore u haysan furayaasha gelitaanka ee qalabku taageero waxay ka heli karaan qiimo gaar ah(ku furmaa daaqad cusub) lammaanahayaga, Yubico. Waxaan sidoo kale qaadeynaa tallaabooyin dheeraad ah si aan u xaddidno gelitaanka hay’adaha khatarta sare leh iyo maamullada sharciyeed ee khatarta sare leh.
ExploitBench: Dhisidda exploit‑yo V8 oo si tartiib-tartiib ah u sii awood badan; GPT‑5.6 wuxuu muujinayaa horumar weyn marka loo eego GPT‑5.5. Jaantuska daahitaanka lama muujinayo maadaama qiyaasidda daahitaanku aanay la isku halayn karin tijaabadan cabbirka waxqabadka.
GPT‑5.6 Sol sidoo kale wuxuu muujinayaa horumar ballaaran guud ahaan cilmi-baarista sayniska. Marka laga hadlayo qiimaynnada culuumta nolosha, GPT‑5.6 wuxuu muujinayaa horumarro Pareto ah marka loo eego GPT‑5.5 oo ku saabsan bayoolojiga dunida dhabta ah, habraacyada shaqo ee cilmi-baarista culuumta nolosha, iyo kiimikada.
GeneBench Pro: Falanqaynta Falanqaynta hidde-sideyaasha iyo iyo bayoolajiga tirada ku saleysan ee muddada dheer; GPT‑5.6 wuxuu gaaraa natiijooyin aad u wanaagsan isagoo adeegsanaya token-no aad u yar iyo waqti kooban. Claude Fable 5 laguma darin maadaama uusan ka jawaabin(ku furmaa daaqad cusub) su'aalaha bayoolojiga heerka sare ah, isla markaana uu diido inta badan su'aalaha ku jira qiimayntan.
GPT‑5.6 waa moodallkeeni ugu awoodda badnaa ilaa hadda ee loogu talagalay dardargelinta cilmi-baarista AI. Gudaha OpenAI, cilmi-baarayaashu waxay u adeegsadaan dhammaan wareegga horumarinta: baarista cilladaha, tayaynta nidaamyada tababarka, samaynta tijaabooyinka, iyo fasiraadda natiijooyinka. Waxaan hore u aragnay dardargelintaas iyo qaadasho sii xooggan intii lagu jiray muddadii tijaabada gudaha ee GPT‑5.6, maadaama celceliska token-nada wax-soo-saarka maalinlaha ah ee cilmi-baare kasta oo firfircoon uu ka badnaa laba jibbaar heerkii ugu sarreeyay ee lagu arkay GPT‑5.5.
Habkan loo shaqeeyo ayaa si dhakhso ah u noqonaya mid caadi ah. Lixdii bilood ee la soo dhaafay, saamiga awoodda xisaabinta ee cilmi-baarista ee loo qoondeeyay saadaasha koodhka gudaha ayaa kordhay 100 jeer, halka isticmaalka token agentic ee gudaha uu kordhay qiyaastii 22 jeer. Cabbirradan qaadashada keligood ma cabbiraan horumarka cilmi-baarista, balse waxay muujinayaan sida degdegga ah ee kaalmada AI ugu sii kordhayso cilmi-baarista iyo guud ahaan kooxaha kale sida iibka, suuqgeynta, hawlgallada isticmaalayaasha, maaliyadda, iyo kuwo kale.
Si aan si toos ah u cabbirno kartidan, waxaan horumarinay xirmo gudaha ah oo qiimeynno ah oo ku saleysan hawlo cilmi-baaris AI oo dhab ah, oo ay ku jiraan saxidda cilladaha nidaamyada cilmi-baarista, hagaajinta kernels iyo hab-raacyada tababarka, fulinta tijaabooyinka barashada mashiinka, iyo hagaajinta nooc kale.
Awoodda guud ee RSI: Marka la eego dhowr imtixaan oo lagu cabbirayo horumarka laga gaaray is-hagaajinta iskeed ah, waxaan aragnaa in GPT‑5.6 Sol uu GPT‑5.5 kaga sarreeyo horumar dhan 16.2 dhibcood, isagoo guud ahaan dardargelinaya cilmi-baarista gudaha.
Marka awoodaha moodalladu kordhaan, waxaan xoojinnaa xirmadayada badbaadada si sirdoonka horumarsan uu weli u ahaado mid si ballaaran waxtar u leh, annaga oo baaritaan dheeraad ah ku samaynayna adeegsiyada khatarta ugu sarreysa. GPT‑5.6, waxaan u dhisnay nidaamkayagii badbaado ee ugu adkaa ilaa hadda, kaas oo lagu waafajiyay awoodaha nooc kasta laguna xoojiyay awood xisaabeed ka badan abid hore.
Moodallada GPT‑5.6 way ka awood badan yihiin moodalladeenii hore dhanka bayoolojiga iyo amniga internetka labadaba, laakiin kama gudbaan xadka Critical midkoodna labada qaybood. Dhanka amniga internetka, tijaabooyinkayagu waxay muujinayaan in GPT‑5.6 uu kaga fiican yahay helidda iyo hagaajinta nuglaanshaha marka loo eego inuu si la isku halleyn karo u fuliyo weerarro iskood u socda oo dhammaad-ilaa-dhammaad ah oo ka dhan ah bartilmaameedyo si adag loo adkeeyay—taasoo difaacayaasha siinaysa fursad ay ku xoojiyaan nidaamyada ka hor inta aan daciifnimada laga faa’iidaysan. Dhanka bayoolojiga, tijaabooyinkayagu waxay muujinayaan in GPT‑5.6 uu taageeri karo cilmi-baaris sharci ah, balse aanu bixin awoodda dhammaystiran ee loo baahan yahay si loo abuuro, loo injineeriyo, ama loo isku-dhiso khatar cusub oo aad u halis badan.
Labada gooboodba asal ahaan waxay leeyihiin adeegsi laba-geesood ah. Dhanka amniga saybarka, awoodaha isku midka ah ee ka caawin kara weeraryahan inuu ka faa’iidaysto nuglaansho ayaa sidoo kale ka caawin kara difaace inuu helo, dib u abuuro, oo uu sameeyo hagaajin la isku halayn karo. Sidaa darteed, xannibaadda xad-dhaafka ahi waxay abuurtaa khatar amni oo u gaar ah. Waxay ka hor istaagi kartaa difaacayaasha inay tijaabiyaan nidaamyada oo ay hirgeliyaan balastarrada, halka jilayaasha waxyeellada leh ay sii wadaan isticmaalka moodallo kale, oo ay ku jiraan moodallada dadwaynaha u daahfuran oo awooddoodu sii kordhayso, iyo sidoo kale agab la aasaasay. Tallaabooyinka ilaalinta ee wax ku oolka ahi waxay tixgeliyaan macnaha guud iyo cawaaqibta ugu badan ee ka dhalan karta codsi, iyagoo ilaalinaya shaqada difaaca ee sharciyeysan halka ay ku dabaqayaan xakameynno ka adag marka caddayntu muujiso khatar culus oo waxyeello.
Tallaabooyinka ilaalinta ee GPT‑5.6 waxaa loo habeeyey lakabyo si loo helo saxnaan iyo kaabis dheeraad ah, waxaana loo naqshadeeyey inay si dhakhso ah ula qabsadaan marka weerarro cusub soo baxaan. Ilaalinnada lagu tababaray moodalka waxay la shaqeeyaan hubinnada waqtiga-dhabta ah, kormeerka joogtada ah, iyo dhaqan-gelinta heer-akooneed, si ay nidaamka uga caawiyaan inuu ammaan ahaado xataa marka lakab gaar ahi uusan u shaqayn sidii loogu talagalay. Nidaamyo badan, calaamadaha kala-soocaha oo keliya ayaa go'aamiya waxa la xannibayo, iyadoo lagu tiirsan yahay moodallo caqli hoose leh oo ay adag tahay in la beddelo si looga hortago waxyeello. Habkeennu wuxuu ku daraa kormeere caqliyeynta oo dib u eega wada sheekaysiga si loo go'aamiyo haddii ay jirto suurtagalnimo waxyeello. Naqshaddan waxaa loogu talagalay inay suurtogeliso shaqada difaaca iyadoo la xannibayo si xun u adeegsiga halista ah, iyadoo awoodaha ugu xasaasisan lagu koobo isticmaaleyaasha la xaqiijiyay iyada oo loo marayo Trusted Access. Maadaama ilaalinnada qaarkood ay adeegsadaan caqliyeynta waqtiga tijaabada, si degdeg ah ayaan u cusboonaysiin karnaa si aan u xirno goldaloolooyinka annaga oo aan dib u tababarin kala-soocayaasha laga bilaabo eber.
Waxaan qaadanaynaa hab aad u taxaddar badan inta aan sii wadno xoojinta nidaamka si looga difaaco weerarrada la-qabsiga leh. Marka la barbar dhigo moodalladii hore, GPT‑5.6-keena Ilaalinnada sayberka ee Sol waxay xannibaan qiyaastii toban jeer hawlo waxyeello leh oo suurtagal ah. Maadaama tallaabooyinkani ay carqalad ku abuuri karaan isticmaalka aan waxyeellada lahayn, waxaan ku bixinnaa ChatGPT iyo Codex ikhtiyaar si fudud loogu tijaabiyo mar kale weydiinta moodallada awood-hoose leh, waxaanan sii wadi doonnaa yareynta saameynta tallaabooyinkayaga ilaalinta ku leedahay isticmaalka aan waxyeellada lahayn annagoo ilaalinayna halbeeg sare oo adkaysi ah. Tani waxay ka tarjumaysaa habkeenna hirgelinta ee ku dhisan tijaabo iyo hagaajin joogto ah: in si taxaddar leh loo bilaabo, kadibna la horumariyo iyadoo lagu salaynayo waxa aan ka baranno isticmaalka dhabta ah.
Ka hor helitaanka guud, waxaan sameynay qiimeynihii badbaadada ugu xoogga badnaa ilaa hadda, oo ay ku jireen qiimeyn ballaaran oo qiimeynta awoodda nidaamka, tijaabooyin adag oo ku saabsan awoodaha iyo ilaalinnada oo lala sameeyay khubaro dibadeed, iyo ku dhawaad 700,000 saacadood oo A100e GPU ah oo loogu talagalay qiimeynta awoodda nidaamka otomaatigga ah ee nooca sanduuqa-madow. Tani waxay noo suuragelisay inaan si nidaamsan u baarno meelaha nuglaanta leh ee suurtagalka ah, soo bandhigno isku dayada jebinta xannibaadaha (jailbreaks), isla markaana naga caawiso inaan nidaamka adkayno ka hor daahfurka.
Ma jiro wax la yiraahdo amni kaamil ah, shaqadayaduna ku aaddan sugidda amniga moodallada awooddoodu sii kordhayso way socotaa. Nuglaanshooyin cusub ayaa la ogaan doonaa, sidoo kalena waxaa jiri doona jebinta xannibaadaha cusub oo ku hareer mara tallaabooyinka ilaalinta ee hadda jira. Jiil kasta oo cusub oo moodal ah wuxuu sidoo kale abuuri doonaa dariiqooyin cusub oo weerar iyo si xun u adeegsi ah. Waxaan u dhisnaa xaqiiqadaas annagoo adeegsanayna tallaabooyin ilaalin oo lakabaysan, kormeer joogto ah, sixid degdeg ah, iyo iskaashi ka dhex jira bulshada difaaca. GPT‑5.6, waxaan barnaamijyadayada amniga ee hore u jiray(ku furmaa daaqad cusub) iyo Barnaamijka Bug Bounty ee bayoolojiga ku lammaaninay hab cusub oo sixitaan degdeg ah iyo dadaalkayagii kormeer ee ugu xoogga badnaa ilaa hadda. Natiijooyinka ka yimaada cilmi-baarayaasha, kormeerka, iyo isticmaalka khaldan ee ka dhaca dunida dhabta ah ayaa si joogto ah loogu dari doonaa qiimayno cusub iyo tallaabooyin ilaalin oo aad u adag.
Akhri wax badan oo ku saabsan ilaalintayada ku jirta kaarka siistamka GPT‑5.6 ee la cusboonaysiiyay(ku furmaa daaqad cusub).
GPT‑5.6 waxa uu ka kooban yahay saddex heerar nooc: Sol, noocayaga hormuudka ah; Terra, nooc qiimo jaban leh oo waxqabadkiisu la tartami karo GPT‑5.5; iyo Luna, noocayaga ugu dhakhsaha badan uguna qiimo-jaban. Lambarku wuxuu tilmaamayaa jiilka, halka Sol, Terra, iyo Luna ay yihiin heerar awood oo waara oo horumari kara jadwalkooda u gaarka ah.
GPT‑5.6 waxaa laga heli karaa laga bilaabo maanta dhammaan ChatGPT, Codex, iyo API-ga OpenAI. Daahfurku wuxuu hadda ka bilaabanayaa caalamka oo dhan, wuxuuna si tartiib tartiib ah u socon doonaa ilaa helitaan buuxa 24-ka saacadood ee soo socda gudahood.
- Sheekaysi: Isticmaalayaasha Plus, Pro, Business, iyo Enterprise waxay GPT‑5.6 Sol marin u helaan dejimaha dadaalka heerka dhexdhexaadka ah iyo kuwa ka sarreeya. Isticmaalayaasha Pro iyo Enterprise sidoo kale waxay dooran karaan GPT‑5.6 Sol Pro si ay u helaan natiijooyinka tayada ugu sarreysa ee hawlaha adag.
- ChatGPT Work iyo Codex: Isticmaalayaasha Free iyo Go waxay galaangal u leeyihiin GPT‑5.6 Terra. Isticmaalayaasha Plus, Pro, Business, iyo Enterprise waxay kala dooran karaan GPT‑5.6 Sol, Terra, iyo Luna oo u deji heerka dadaalka mid kasta.
maxwaxaa heli kara dhammaan isticmaalayaasha marin u leh GPT‑5.6 ee ChatGPT Work iyo Codex, waxaana laga daaran karaa dejimaha. Gudaha ChatGPT Work,ultrawaxaa heli kara isticmaalayaasha Pro iyo Enterprise. Gudaha Codex, waxa laga heli karaa qorshaha Plus iyo qorshayaasha ka sarreeya. - API: Horumariyeyaashu waxay geli karaan Sol, Terra, iyo Luna iyagoo adeegsanaya OpenAI API. Gudaha Responses API-ga, Wacitaanka Qalabka ee Barnaamijeysan wuxuu GPT‑5.6 u oggolaanayaa inuu qoro oo ku socodsiiyo barnaamijyo gudaha xusuusta, kuwaas oo isku dubbarida qalabka isla markaana farsameeya natiijooyinka dhexe, taasoo ka dhigaysa mid la jaanqaadi karta Xog Haynta Eberka ah (ZDR). Multi-agent, oo markii hore lagu heli karo nooca beta, wuxuu GPT‑5.6 u oggolaanayaa inuu isku mar socodsiiyo wakiillo-hoosaadyo, shaqadoodana ku mideeyo hal codsi.
GPT‑5.6 waxaa lagu qiimeeyaa halkii 1M shay iyadoo lagu saleeyay saddex cabbir oo moodal ah: Sol waa $5 gelin / $30 soo-saar; Terra waa $2.50 gelin / $15 soo-saar; Luna-na waa $1 gelin / $6 soo-saar. GPT‑5.6 sidoo kale wuxuu soo bandhigayaa kaydinta weydiimaha oo la sii saadaalin karo, oo ay ku jirto taageerada meelaha si cad loogu jaro kaydka(ku furmaa daaqad cusub) iyo muddada ugu yar ee kaydku jiro oo ah 30 daqiiqo. GPT‑5.6 iyo moodallada ka dambeeya, qorista kaydka waxaa lagu dallacaa 1.25x heerka gelinta aan kaydsanayn ee moodalka, halka akhriska kaydka uu sii wato helidda dhimista 90% ee gelinta kaydsan.
Mihnad
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Fable 5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview | Gemini 3.5 Flash |
| Tijaabadii Ugu Dambeeysay ee Wakiillada | 52.7% | 50.4% | 50.3% | 46.9% | 40.5% | 45.2% | 32.1% | — |
| GDPval-AA v2 | 1,747.8 Elo | 1,593 Elo | 1,591.8 Elo | 1,493.7 Elo | 1,759.6 Elo | 1,600.1 Elo | 962.3 Elo | 1,348.8 Elo |
| Hawlaha La-talinta Maareynta (Gudaha) | 43.2% | 37.2% | 35.4% | 31.3% | 35.5% | 31.6% | 13.2% | — |
| Tijaabada Maaliyadda Weyn | 53% | 51% | 36% | 49% | — | 44% | — | — |
| Tusmada Sirdoonka ee Artificial Analysis v4.1 | 58.9 Dhibcaha guud | 55 Dhibcaha guud | 51.2 Dhibcaha guud | 54.8 Dhibcaha guud | 59.9 Dhibcaha guud | 55.7 Dhibcaha guud | 46.5 Dhibcaha guud | 50.2 Dhibcaha guud |
Qorista barnaamijyada kombiyuutarka
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Sol Ultra | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Mythos 5 | Claude Mythos Preview | Claude Fable 5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview |
| Artificial Analysis Coding Agent Index v1.1 | 80 Dhibcaha guud | — | 77.4 Dhibcaha guud | 74.6 Dhibcaha guud | 76.4 Dhicaha guud | — | — | 77.2 Dhibcaha guud | 72.5 Dhibcaha guud | 42.7 Dhibcaha guud |
| SWE-Bench Pro | 64.6% | — | 63.4% | 62.7% | 59.4% | 80.3% | 77.8% | 80% | 69.2% | 54.2% |
| DeepSWE v1.1 | 72.7% | — | 69.6% | 67.2% | 67% | — | — | 69.7% | 59% | 11.8% |
| Terminal-Bench 2.1 | 88.8% | 91.9% | 87.4% | 84.7% | 85.6% | 88% | — | 83.1% | 78.9% | 70.7% |
Badqabka
| Qiimayn | GPT‑5.6 Xal | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | GPT‑5.4 | Claude Opus 4.8 | Claude Mythos 5 | Claude Mythos Preview |
| Healthbench Professional | 60.5% | 57.7% | 55.7% | 51.8% | 48.1% | 52.6% | 66% | 64.7% |
Isticmaalka kombiyuutarka
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Sol Ultra | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Mythos 5 | Claude Mythos Preview | Claude Opus 4.8 | Gemini 3.1 Pro Preview |
| OSWorld 2.0 | 62.6% | — | 50.2% | 45.6% | 47.5% | — | — | 54.8% | — |
| BrowseComp | 90.4% | 92.2% | 87.5% | 83.3% | 84.4% | 88% | 87.9% | 84.3% | 85.9% |
| BenchCAD | 70.6% | — | 62.3% | 63.1% | 44.4% | 38.4% | 35.5% | 27.3% | — |
| BenchCAD (qalab Python) | 83.4% | — | 78.2% | 73.9% | 55.8% | 65% | 61% | 51.8% | — |
Saybar
| Qiimaynta | GPT‑5.6 Xalka | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.6 Sol MA | GPT‑5.5 | GPT‑5.4 | Claude Opus 4.8 | Claude Opus 4.7 | Claude Mythos 5 | Claude Mythos Preview | Claude Opus 4.6 |
| CVEBench | 94.1% | 95.1% | 91.2% | — | 94.1% | 92.2% | — | — | — | — | — |
| HoudiniBench | 0% | 0% | 0% | — | — | — | — | — | — | — | — |
| Hawlaha tartanka Capture-the-Flags (Gudaha) | 96.7% | 91.8% | 85.2% | — | 88.1% | 83.7% | — | — | — | — | — |
| SEC-bench Pro | 71.2% | 57.7% | 48.9% | 76.2% | 45.8% | 35.9% | — | — | — | — | — |
| Cybergym | 84.5% | 81.8% | 77.9% | — | 81.8% | 79% | 78.1% | 73% | 83.8% | 83% | — |
| EVMBench Detect | 66.9% | 71.3% | 64.4% | — | — | — | — | — | — | — | — |
| EVMBench Ka-faa’iidaysi | 20.3% | 21.9% | 18.8% | — | — | — | — | — | — | — | — |
| ExploitGym (port gudaha ah) | 63.5% | 44.3% | 16% | — | 25.9% | 12.7% | — | — | — | — | — |
| VulnLMP (Qiimaynta PrepX) | — | — | — | — | — | — | — | — | — | — | — |
| ExploitBench | 73.5% | 52.9% | 33.2% | — | 47.9% | 38% | 40% | 27.7% | 78% | 74.2% | — |
| ExploitGym | 33.7% | 23.2% | 12.4% | — | 15.1% | 7% | — | — | — | — | 1.8% |
Is-horumarin
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 |
| Qiimaynta Cillad-saarista Cilmi-baarista Gudaha | 68.3% | 67.8% | 50.8% | 50% |
| KernelGen 1P | 61.1% | 49.2% | 22.4% | 29.3% |
| NanoGPT | 9.69% | 14.5% | 1.66% | 2.65% |
| PostTrainBench Lite | 50.3% | 51.5% | 29.6% | 38.8% |
| Tusmada RSI | 57.9% | 56.3% | 41.9% | 41.7% |
Qaabab badan leh
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Fable 5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview |
| MMMU Pro (qalab la’aan) | 83% | 80.7% | 78.4% | 81.2% | — | — | 80.5% |
| MMMU Pro (qalab leh) | 84.6% | 82% | 79.5% | 83.2% | — | — | — |
| gdp.pdf | 30.7% | 24.7% | 22.7% | 26% | 29.8% | 22.5% | 16.7% |
Waxbarasho
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Mythos 5 | Claude Mythos Preview | Claude Fable 5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview |
| GPQA Diamond | 94.6% | 92.9% | 92.3% | 93.6% | 94.1% | 94.6% | 92.6% | 92% | 94.3% |
| FrontierMath Heerarka 1-3 (v2) | 89% | 84.9% | 78.6% | 85.3% | — | — | 87% | 80% | 59.6% |
| FrontierMath Heerka 4 (v2) | 83% | 68.3% | 58.5% | 72.5% | — | — | 87.8% | 56.1% | — |
Isticmaalka qalabka
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Mythos 5 | Claude Mythos Preview | Claude Fable 5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview | Gemini 3.5 Flash |
| AutomationBench | 18.1% | 15.2% | 14.9% | 12.9% | — | — | 17.4% | 15.5% | — | 14.5% |
| Toolathlon | 58% | 53.1% | 53.4% | 55.6% | 61.7% | 61.1% | 61.7% | 59.9% | 48.8% | — |
Macnaha guud ee dheer
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Mythos 5 | Claude Mythos Preview | Claude Opus 4.8 |
| OpenAI MRCR v2 8-needle 256K-512K | 91.5% | 89.6% | 41.3% | 81.5% | — | — | — |
| OpenAI MRCR v2 8-needle 512K-1M | 73.8% | 72.5% | 41.3% | 74% | — | — | — |
| GraphWalks BFS 256k f1 | 90.7% | 76.9% | 81.3% | 73.7% | 91.1% | 85.7% | 85.9% |
| GraphWalks BFS 1mil f1 | 77.1% | 71.2% | 51.2% | 45.4% | 79.4% | 74.3% | 68.1% |
Caqliyeynta aan la taaban karin
| Qiimaynta | GPT‑5.6 Sol | GPT‑5.6 Terra | GPT‑5.6 Luna | GPT‑5.5 | Claude Opus 4.8 | Gemini 3.1 Pro Preview |
| ARC-AGI-3⁷ | 7.78% | 0.8% | 0.18% | 0.43% | 1.5% | 0.42% |
Qoraa
Qoraallada hoose
1. Awoodaha sayberka waxaa lagu qiimeeyaa iyadoo la adeegsanayo tallaabooyin ilaalin oo la dhimay. Isticmaalayaashu waxay ku biiri karaan barnaamijka Trusted Access for Cyber ee OpenAI Daybreak si ay u helaan gelitaan dheeraad ah oo ku saabsan awoodaha difaaca saybarka.
2. Dhammaan nooc-yada waxaa lagu qiimeeyaa iyadoo la adeegsanayo ExploitBench API harness oo leh 5 setinno aan kala sooc lahayn iyo joogteynta caqliyeynta.
3. Waxaan ku socodsiinnay ExploitGym API-gayaga noociisa hore, kaas oo soo saara jawaabo ka dheereeya API-gayaga dadweynaha, dabadeedna waan dib-u-miisaannay si uu ula jaanqaado API-gayaga dadweynaha. Marka dib-u-dhacyada dib loogu miisaamo xawaareyaasha la filayo ee API-gayaga dadweynaha, tani waxay keentaa in qaar ka mid ah dib-u-dhacyada la qiyaasay ay ka bataan xadka waqtiga ee laba- iyo lix-saacadood, inkastoo si sax ah loo raacay intii lagu jiray orodka qiimeynta. Si loo helo xawaare degdeg ah shaqooyinka waqtiga xasaasiga ah, waxaan ku bixinnaa habayn mudnaan leh gudaha API-ga iyo hab degdeg ah gudaha Codex.
4. Waxaan qiyaasnaa dib-u-dhaca iyo kharashka API-ga annagoo eegayna hab-dhaqanka wax-soo-saarka ee moodalladeenna, kuna samaynayna jilid ofleen ah. Qiyaasahani waxay tixgeliyaan faahfaahinta wicitaanka qalabka, qeybo-qoraallada la muunadeeyay, iyo qeybo-qoraallada gelinta. Natiijooyinka dunida dhabta ah aad bay u kala duwanaan karaan, waxayna ku xiran yihiin arrimo badan oo aan ku jirin jiliddeena. Waxaan jilnaa dib-u-dhaca xawaareyaal API oo degdeg ah, kharashkana qiimeynta API ee caadiga ah.
5. Moodallada aan laga soo sheegin token-nada soo-saarka, daahitaanka ama kharashka waxaa lagu muujinayaa xariiqyo dhibco leh oo jiif ah.
6. Marka ay tahay wakiillo badan, daahitaanku wuxuu ka yimaadaa wakiilka asalka ah, halka wadarta token-nada wax-soo-saarka iyo kharashka API-ga ay ku jiraan dhammaan token-nada. Ultra waxaa lagu socodsiiyaa 4 wakiil.
7. Waxaan dhibcaha ku xisaabinnaa habka rasmiga ah ee dhibcaynta ee lagu sharraxay warqadda HealthBench Professional, kaas oo aan la barbar dhigi karin natiijooyinka lagu sheegay kaarka siistamka ee Anthropic.
8. ARC-AGI-3 ee Opus 4.8 waxaa lagu socodsiiyay dadaalka caqliyeynta oo sarreeya, balse aan ahayn kan ugu sarreeya, maadaama tani ay tahay natiijada keliya ee ARC-AGI-3 ee la daabacay.
