Cusboonaysiin Abriil 24, 2026: GPT‑5.5 iyo GPT‑5.5 Pro hadda waxa laga heli karaa API-ga. Kaarka siistamka sidoo kale waa la cusboonaysiiyay si loo sharaxo ilaalinnada dheeraadka ah ee khuseeya.
Waxaan soo saaraynaa GPT‑5.5, noockeenna ugu caqli badan uguna dareen-fudud isticmaalka ilaa hadda, waana tallaabada xigta ee ku wajahan hab cusub oo shaqo loogu qabto kombiyuutarka.
GPT‑5.5 si ka dhakhso badan ayay u fahantaa waxa aad isku dayayso inaad qabato, waxayna qaadi kartaa qayb badan oo shaqada ah lafteeda. Waxay ku fiican tahay qorista iyo sixidda koodhka, cilmi-baarista onlaynka, falanqaynta xogta, abuurista dukumeenti iyo xaashiyo-xogeed, ku shaqaynta software, iyo u kala gudbidda qalabyo kala duwan ilaa hawshu dhammaato. Halkii aad si taxaddar leh u maamuli lahayd tallaabo kasta, waxaad GPT‑5.5 siin kartaa hawl qasan oo qaybo badan leh, kuna aamini kartaa inay qorshayso, qalab isticmaasho, shaqadeeda hubiso, mugdiga dhex marto, oo sii waddo.
Horumarradu si gaar ah ayay ugu xooggan yihiin coding-ka agentic, isticmaalka kombiyuutarka, shaqada aqoonta ku salaysan, iyo cilmi-baarista sayniska ee hore—meelo ay horumarku ku xiran yihiin caqliyeynta guud ahaan macnaha iyo tallaabo-qaadista waqti ka dib. GPT‑5.5 waxay keentaa kororkan caqliga ah iyadoo aan xawaaraha loo hurin: noocyada waaweyn ee awoodda badan badanaa way gaabiyaan bixinta, balse GPT‑5.5 waxay la jaanqaaddaa GPT‑5.4 marka la eego dib-u-dhaca token-kiiba ee adeeg-bixinta dunida dhabta ah, iyada oo qabanaysa heer caqli aad uga sarreeya. Waxay sidoo kale isticmaashaa tokenno si weyn uga yar si ay u dhammaystirto isla hawlaha Codex, taas oo ka dhigaysa mid ka hufan iyo sidoo kale mid awood badan.
Waxaan GPT‑5.5 ku sii daynaynaa xirmadii ilaalin ee noogu adkayd ilaa maanta, oo loo naqshadeeyay in lagu yareeyo si-xun-u-isticmaalka iyadoo la ilaalinayo helitaanka shaqada faa’iidada leh. Waxaan noocan ku qiimaynay dhammaan xirmooyinkeenna badbaado iyo Qaab-dhismeedka Diyaar-garowga, waxaan la shaqaynay redteamers gudaha iyo dibadda ah, waxaan ku darnay tijaabooyin bartilmaameed u ah awoodaha horumarsan ee amniga internetka iyo bayoolajiga, waxaanan ka ururinnay jawaab-celin ku saabsan kiisaska isticmaalka dhabta ah ku dhowaad 200 oo la-hawlgalayaal hore u helay ka hor sii-daynta.
Maanta, GPT‑5.5 waxa loo soo bandhigayaa isticmaalayaasha Plus, Pro, Business, iyo Enterprise ee ChatGPT iyo Codex, halka GPT‑5.5 Pro loo soo bandhigayo isticmaalayaasha Pro, Business, iyo Enterprise ee ChatGPT. Hirgelinnada API-gu waxay u baahan yihiin ilaalinno kala duwan, waxaana si dhow ula shaqaynaynaa la-hawlgalayaasha iyo macaamiisha si loo daboolo shuruudaha badbaado iyo amni ee lagu adeegayo cabbir ballaaran. Waxaan GPT‑5.5 iyo GPT‑5.5 Pro u keenaynaa API-ga si aad ah dhow.
GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro | |
Terminal-Bench 2.0 | 82.7% | 75.1% | - | - | 69.4% | 68.5% |
Expert-SWE (Gudaha) | 73.1% | 68.5% | - | - | - | - |
GDPval (guulo ama barbaro) | 84.9% | 83.0% | 82.3% | 82.0% | 80.3% | 67.3% |
OSWorld-Verified | 78.7% | 75.0% | - | - | 78.0% | - |
Toolathlon | 55.6% | 54.6% | - | - | - | 48.8% |
BrowseComp | 84.4% | 82.7% | 90.1% | 89.3% | 79.3% | 85.9% |
FrontierMath Tier 1–3 | 51.7% | 47.6% | 52.4% | 50.0% | 43.8% | 36.9% |
FrontierMath Tier 4 | 35.4% | 27.1% | 39.6% | 38.0% | 22.9% | 16.7% |
CyberGym | 81.8% | 79.0% | - | - | 73.1% | - |
OpenAI waxay dhisaysaa kaabeyaasha caalamiga ah ee AI-ga agentic, taas oo u suuragelinaysa dadka iyo ganacsiyada dunida oo dhan inay shaqo ku qabtaan AI. Sanadkii la soo dhaafay, waxaan aragnay AI oo si weyn u dardargelinaysa injineernimada software-ka. Iyadoo GPT‑5.5 ku jirto Codex iyo ChatGPT, isbeddelkaas oo kale wuxuu hadda bilaabmayaa inuu ku fido cilmi-baarista sayniska iyo guud ahaan shaqada ay dadku ku qabtaan kombiyuutarrada.
Guud ahaan dhinacyadan, GPT‑5.5 ma aha oo kaliya mid ka caqli badan; waa mid ka hufan sida ay dhibaatooyinka uga shaqayso, iyadoo badanaa gaadha wax-soo-saar tayo sare leh tokenno yar iyo dib-u-dayisyo yar. Tusmada Coding-ka ee Artificial Analysis, GPT‑5.5 waxay keentaa caqli heerka ugu sarreeya ah iyadoo qiimuhu yahay kala bar kan noocyada coding-ka ugu casriyeysan ee tartanka ah.
Tusmada Sirdoonka Artificial Analysis(ku furmaa daaqad cusub) waa celcelis miisaan leh oo ka kooban 10 qiimeyn oo ay maamulaan dhinac dibadeed: AA-LCR, AA-Omniscience, CritPt, GDPval-AA, GPQA Diamond, Humanity’s Last Exam, IFBench, SciCode, Terminal-Bench Hard, τ²-Bench Telecom.
GPT‑5.5 waa noockeenna coding-ka agentic ugu xooggan ilaa maanta. Marka la eego Terminal-Bench 2.0, oo tijaabiya qulqullo shaqo oo adag oo khadka amarada ah kuwaas oo u baahan qorshayn, ku-celcelin, iyo isuduwidda qalabka, waxay gaadhaa saxnaan heerka ugu sarreeya ah oo ah 82.7%. Marka la eego SWE-Bench Pro, oo qiimeeya xallinta arrimaha GitHub ee dunida dhabta ah, waxay gaadhaa 58.6%, iyadoo xallinaysa hawlo badan dhammaad-ilaa-dhammaad ah hal mar oo keliya marka loo eego noocyadii hore. Marka la eego Expert-SWE, qiimaynteena gudaha ee ugu casriyeysan ee hawlaha coding-ka aragti-dheer leh oo leh waqtiga dhamaystirka aadanaha ee la qiyaasay ee dhexdhexaadka ahi yahay 20 saacadood, GPT‑5.5 sidoo kale way ka sarreysaa GPT‑5.4.
Saddexdan qiimayn oo dhan, GPT‑5.5 waxay ka fiicnaataa dhibcaha GPT‑5.4 iyadoo isticmaalaysa tokenno yar.
Awoodaha coding-ka ee noocku waxay si gaar ah ugu muuqdaan Codex halkaas oo ay qaadi karto shaqooyin injineernimo oo u dhexeeya hirgelin iyo dib-u-habayn ilaa sixid, tijaabin, iyo xaqiijin. Tijaabinta hore waxay muujinaysaa in GPT‑5.5 ay ka fiican tahay dabeecadaha ay shaqada injineernimada dhabta ahi ku tiirsan tahay, sida haynta macnaha guud ee nidaamyada waaweyn, caqliyeynta fashilaadaha mugdiga leh, hubinta mala-awaallada iyadoo qalab la adeegsanayo, iyo gudbinta isbeddellada inta ka hartay saldhigga koodhka.
Jidka la muujiye wuxuu adeegsadaa xogta vector-ka ee NASA/JPL Horizons ee Orion, Dayaxa, iyo Qorraxda, iyadoo la mariyey qiyaas-muujin si ay u fududaato akhrisku.
Weydiin: [sawir lifaaqan] Tan u hirgeli sidii app cusub adigoo adeegsanaya webgl iyo vite kuna saleynaya xogta dhabta ah ee howlgalka Artemis II. Hubi inaad si fiican u tijaabiso app-ka ilaa uu si buuxda u shaqeynayo oo uu u ekaado app-ka sawirka ku jira. Si dhow ugu fiirso sawiridda meerayaasha iyo waddooyinka duulista. Waxaan rabaa inaan la falgeli karo muuqaalka 3D-ga. Hubi inuu leeyahay makaanikada wareegga xiddigaha ee dhab ah.
Marka laga gudbo bar-bardhigyada, tijaabiyeyaashii hore waxay sheegeen in GPT‑5.5 ay muujinayso awood ka xooggan oo ah fahamka qaabka nidaamka: sababta wax u fashilmayaan, meesha sixitaanku ku habboon yahay, iyo waxa kale ee saldhigga koodhka ku saameyn doona.
“Noockii ugu horreeyay ee kood qorista ee aan isticmaalo oo leh caddeyn fikradeed oo dhab ah.”
Dan Shipper, Aasaasaha iyo CEO-ga Every, wuxuu GPT‑5.5 ku tilmaamay “noockii ugu horreeyay ee kood qorista ee aan isticmaalo oo leh caddeyn fikradeed oo dhab ah.”
Ka dib markii uu bilaabay app, wuxuu maalmo ku qaatay sixidda dhibaato soo baxday bilowga ka dib ka hor inta uusan keenin mid ka mid ah injineerradiisa ugu fiican si uu dib ugu qoro qayb ka mid ah nidaamka. Si loo tijaabiyo GPT‑5.5, wuxuu si wax-ku-ool ah dib ugu celiyay saacadda: ma eegi kartaa noocku xaaladda jabtay oo ma soo saari kartaa isla nooca dib-u-qorista ee injineerku ugu dambayn go’aansaday? GPT‑5.4 ma aysan karin. GPT‑5.5 way kartay.
“Runtii waxay ila tahay inaan la shaqaynayo sirdoon ka sarreeya, waxaana ku dhow dareen xushmad leh.”
Pietro Schirano, CEO-ga MagicPath, wuxuu arkay isbeddel tallaabo la mid ah markii GPT‑5.5 uu isku daray branch ay ku jireen boqolaal isbeddel frontend ah iyo dib-u-habeyn branch weyn oo isna si weyn isu beddelay, isagoo shaqada ku xalliyay hal mar ku dhowaad 20 daqiiqo gudahood.
Injineerro sare oo tijaabiyay nooca ayaa sheegay in GPT‑5.5 ay si muuqata uga xoog badnayd GPT‑5.4 iyo Claude Opus 4.7 xagga caqliyeynta iyo madaxbannaanida, iyadoo horay u qabanaysa arrimo iyo saadaalinta baahiyaha tijaabinta iyo dib-u-eegista iyada oo aan si cad loo weydiin. Hal kiis, injineer ayaa ka codsaday inay dib u qaabayso nidaam faallooyin ah oo ku jira tifaftiraha markdown wada-shaqayn ah, wuxuuna ku soo noqday xirmo 12-diff ah oo ku dhowaad dhammaystiran. Qaar kalena waxay sheegeen inay si la yaab leh ugu baahdeen sixid hirgelin oo yar waxayna dareemeen kalsooni badan qorshayaasha GPT‑5.5 marka loo eego GPT‑5.4.
Mid ka mid ah injineerrada NVIDIA ee hore u helay nooca ayaa xitaa gaadhay inuu yidhaahdo: “Luminta helitaanka GPT‑5.5 waxay ii dareemaysaa sidii in addin la iga gooyay.”
“GPT-5.5 is noticeably smarter and more persistent than GPT-5.4, with stronger coding performance and more reliable tool use. It stays on task for significantly longer without stopping early, which matters most for the complex, long-running work our users delegate to Cursor.”
Awoodahaas isla ah ee GPT‑5.5 ka dhigaya mid ku fiican coding-ka ayaa sidoo kale ka dhigaya mid awood badan shaqada maalinlaha ah ee kombiyuutarka. Sababtoo ah noocku wuxuu ka fiican yahay fahamka ujeeddada, wuxuu si dabiici ah uga gudbi karaa wareegga buuxa ee shaqada aqoonta ku salaysan: helitaanka macluumaadka, fahamka waxa muhiimka ah, qalab isticmaalka, hubinta wax-soo-saarka, iyo u rogidda waxyaabaha ceeriin wax faa’iido leh.
Gudaha Codex, GPT‑5.5 way ka fiican tahay GPT‑5.4 abuurista dukumeenti, xaashiyo-xogeed, iyo bandhigyo slide ah. Tijaabiyeyaasha alfa waxay sheegeen inay ka sar martay noocyadii hore shaqooyin sida cilmi-baaris hawlgal, qaabaynta xaashiyaha-xogeed, iyo u rogista xog ganacsi oo qasan qorshayaal. Marka lagu daro xirfadaha isticmaalka kombiyuutarka ee Codex, GPT‑5.5 waxay nagu soo dhowaynaysaa dareenka ah in noocku dhab ahaan kula isticmaali karo kombiyuutarka: aragista waxa shaashadda ku jira, gujinta, qorista, dhexmarista is-dhexgallada, iyo u kala gudbidda qalabyo si sax ah.
Kooxo ka tirsan OpenAI ayaa durba adeegsanaya awoodahan qulqullo shaqo oo dhab ah. Maanta, in ka badan 85% shirkadda ayaa isticmaasha Codex toddobaad kasta hawlo ay ka mid yihiin injineernimada software-ka, maaliyadda, isgaarsiinta, suuq-geynta, cilmiga xogta, iyo maaraynta wax-soo-saarka. Qaybta Comms, kooxdu waxay GPT‑5.5 ku isticmaashay Codex si ay u falanqayso lix bilood oo xog codsiyo hadal ah, u dhisto qaab-dhismeed dhibcoyn iyo khatar ah, una xaqiijiso wakiil Slack oo otomaatig ah si codsiyada khatarta hoose leh si toos ah loo maareeyo halka kuwa khatarta sare leh weli loo gudbiyo dib-u-eegis aadanaha. Qaybta Maaliyadda, kooxdu waxay Codex u isticmaashay dib-u-eegista 24,771 foomam canshuur K-1 ah oo isu geeynaya 71,637 bog, iyadoo adeegsanaysa qulqul-shaqo ka reebay macluumaadka shakhsiyeed oo kooxda ka caawiyay inay shaqada dedejiyaan laba toddobaad marka loo eego sanadkii hore. Kooxda Go-to-Market, shaqaale ayaa otomaatig ka dhigay samaynta warbixinno ganacsi oo toddobaadle ah, taas oo badbaadisay 5-10 saacadood toddobaadkii.
Gudaha ChatGPT, GPT‑5.5 Thinking waxay fureysaa gargaar degdeg ah oo loogu talagalay dhibaatooyin adag, iyadoo leh jawaabo ka caqli badan oo kooban si ay kaaga caawiso inaad si hufan uga gudubto shaqo adag. Waxay ku fiican tahay shaqada xirfadeed sida coding, cilmi-baaris, isu-geynta iyo falanqaynta macluumaadka, iyo hawlaha dukumeentiyada culus, gaar ahaan marka plugins la adeegsanayo.
Gudaha GPT‑5.5 Pro, tijaabiyeyaashii hore waxay arkayaan koror weyn oo xagga adkaanta iyo tayada shaqada ah ee ChatGPT qaadi karto, iyadoo hagaajinta dib-u-dhaca ay ka dhigayso mid aad uga suurtagal badan hawlaha culus. Marka loo eego GPT‑5.4 Pro, tijaabiyeyaashu waxay heleen jawaabaha GPT‑5.5 Pro kuwo si weyn uga dhammaystiran, si fiican u habaysan, sax ah, la xiriira, oo faa’iido leh, gaar ahaan waxqabad xooggan oo dhinacyada ganacsiga, sharciga, waxbarashada, iyo cilmiga xogta ah.
GPT‑5.5 waxay gaadhaa waxqabad heerka ugu sarreeya ah bar-bardhigyo badan oo ka turjumaya noocan shaqo. Marka la eego GDPval, oo tijaabiya awoodda wakiilladu u leeyihiin inay soo saaraan shaqo aqooneed si fiican loo qeexay oo ka kooban 44 xirfadood, GPT‑5.5 waxay keentaa 84.9%. Marka la eego OSWorld-Verified, oo cabbira in nooc awood u leeyahay inuu keligiis ku shaqeeyo jawi kombiyuutar oo dhab ah, waxay gaadhaa 78.7%. Marka la eego Tau2-bench Telecom, oo tijaabiya qulqullo adeeg-macaamiil oo adag, waxay gaadhaa 98.0% iyada oo aan weydiin la hagaajin. GPT‑5.5 sidoo kale waxay si xooggan uga shaqaysaa bar-bardhigyo kale oo shaqo aqooneed: 60.0% marka la eego FinanceAgent, 88.5% marka la eego hawlaha gudaha ee qaabaynta bangiyada maalgashiga, iyo 54.1% marka la eego OfficeQA Pro.
Tau2-bench Telecom waxaa la socodsiiyay iyada oo aan wax prompt tuning ah la samayn (iyo GPT‑4.1 oo ah nooca isticmaalaha). GPT‑5.5 wuxuu si ka fiican kuwii ka horreeyay u fahmaa ujeeddada hawsha wuxuuna ka hufan yahay token ahaan.
“GPT-5.5 wuxuu keenaa waxqabadka joogtada ah ee looga baahan yahay shaqada culus ee fulinta. Waxaa lagu dhisay laguna bixiyay nidaamyada NVIDIA GB200 NVL72, noocku wuxuu u suurageliyaa kooxaheena inay ka soo saaraan astaamo dhammaad-ilaa-dhammaad ah weydiimaha luqadda dabiiciga ah, yareeyaan waqtiga qalad-bixinta maalmo ilaa saacado, oo ay u beddelaan toddobaadyo tijaabo ah horumar habeen keliya ah oo ku jira codebases adag. Waxay ka badan tahay kood degdeg badan—waa hab cusub oo shaqo oo dadka ka caawiya inay ku shaqeeyaan xawaare aasaas ahaan ka duwan.”
GPT‑5.5 sidoo kale waxay muujisaa horumar qulqullo shaqo oo cilmi-baaris saynis iyo farsamo ah, kuwaas oo u baahan wax ka badan ka jawaabista su’aal adag. Cilmi-baarayaashu waxay u baahan yihiin inay fikrad baaraan, caddeyn ururiyaan, mala-awaallo tijaabiyaan, natiijooyinka fasiraan, oo go’aamiyaan waxa xiga ee la isku dayayo. GPT‑5.5 way ka fiican tahay ku sii adkaysiga wareeggaas marka loo eego noocyo kale.
Gaar ahaan, GPT‑5.5 waxay muujisaa horumar cad oo ka sarreeya GPT‑5.4 marka la eego GeneBench(ku furmaa daaqad cusub), oo ah qiimayn cusub oo diiradda saarta falanqaynta xogta sayniska ee marxalado badan leh ee hidde-sideyaasha iyo bayoolajiga tirada. Dhibaatooyinkani waxay ka rabaan noocyada inay ka fikiraan xog suuragal ahaan mugdi leh ama khalad leh iyadoo hagid kormeer oo yar jirto, inay wax ka qabtaan caqabadaha dhabta ah sida qas-wadayaasha qarsoon ama fashilaadaha QC, iyo inay si sax ah u hirgeliyaan una fasiraan hababka tirakoobka casriga ah. Waxqabadka noocan ayaa aad u yaab leh marka la eego xaqiiqda ah in hawlaha halkan ka jira ay badanaa la jaanqaadaan mashaariic dhowr maalmood qaata oo ay qabtaan khubaro saynis ah.
Sidoo kale, marka la eego BixBench(ku furmaa daaqad cusub), oo ah bar-bardhig loo qaabeeyay bioinformatics-ka iyo falanqaynta xogta ee dunida dhabta ah, GPT‑5.5 waxay gaadhay waxqabad hoggaamineed oo ka mid ah noocyada leh dhibco la daabacay. Awoodaha saynis ee noocan hadda waa kuwo ku filan inay si macne leh u dedejiyaan horumarka xuduudaha cilmi-baarista bayomedikalka iyada oo ah wada-saynisyahan dhab ah.
Tusaale kale, nooc gudaha ah oo GPT‑5.5 ah oo wata harness gaar loo habeeyay ayaa gacan ka geystay helitaanka caddayn cusub(ku furmaa daaqad cusub) oo ku saabsan tirooyinka Ramsey, oo ka mid ah walxaha dhexe ee combinatorics. Combinatorics waxay daraasaysaa sida walxaha go’an isugu habboonaanayaan: garaafyo, shabakado, sets, iyo qaabab. Tirooyinka Ramsey waxay si kooban u waydiiyaan inta ay le’eg tahay in shabakaddu noqoto ka hor inta aan nooc amar ahi si dammaanad leh u muuqan. Natiijooyinka aaggan waa dhif, badanaana farsamo ahaan way adag yihiin. Halkan, GPT‑5.5 waxay heshay caddayn xaqiiqo asymptotic ah oo muddo dheer taagnayd oo ku saabsan tirooyinka Ramsey ee off-diagonal, taas oo markii dambe lagu xaqiijiyay Lean. Natiijadu waa tusaale muuqda oo ah GPT‑5.5 oo aan wax ku darsanayn oo kaliya koodh ama sharaxaad, balse dood xisaabeed la yaab leh oo faa’iido leh oo ku jirta aag cilmi-baaris oo sal ah.
Tijaabiyeyaashii hore waxay GPT‑5.5 Pro ku isticmaaleen ChatGPT si ka yar mishiin jawaab hal mar ah oo ka badan lammaane cilmi-baaris: iyagoo naqdiyaya qoraallo cilmiyeed dhowr wareeg, adkayn ku samaynaya doodaha farsamo, soo jeedinaya falanqayn, kana shaqaynaya koodh, qoraallo, iyo macnaha PDF. Xadhigga mideeya ayaa ah in GPT‑5.5 ay ka fiican tahay ka caawinta cilmi-baarayaasha inay ka gudbaan su’aal ilaa tijaabo ilaa wax-soo-saar.
Derya Unutmaz, borofisar iyo cilmi-baare immunology ah oo ka tirsan Jackson Laboratory for Genomic Medicine, wuxuu GPT‑5.5 Pro u adeegsaday falanqaynta dataset muujinta gene-ka ah oo leh 62 muunadood iyo ku dhowaad 28,000 genes, isagoo soo saaray warbixin cilmi-baaris oo faahfaahsan taas oo aan kaliya soo koobayn natiijooyinka balse sidoo kale iftiimisay su’aalo iyo aragtiyo muhiim ah—shaqo uu sheegay inay kooxdiisu bilo ku qaadan lahayd.
Bartosz Naskręcki, kaaliye borofisar xisaabeed ka ah Jaamacadda Adam Mickiewicz ee Poznań, Poland, wuxuu GPT‑5.5 ku isticmaalay Codex si uu uga dhiso app algebraic-geometry ah hal weydiin gudaheed 11 daqiiqo, isagoo muujiya isgoyska oogada quadratic-ka ah una beddelaya qalooca natiijada nooc Weierstrass ah.
Markii dambe wuxuu app-ka ku ballaariyay muuqaalin singularity oo xasilloon iyo coefficients sax ah oo dib loogu adeegsan karo shaqooyin dambe. Isaga, isbeddelka weyn waa in Codex uu hadda ka caawin karo hirgelinta muuqaalinta xisaabeed ee gaar ah iyo socod-shaqooyinka computer-algebra ee hore ugu baahnaa qalab gaar ah. Isku darka, tusaalooyinkani waxay muujinayaan GPT‑5.5 oo u beddelaya ujeeddada khabiirka qalab cilmi-baaris iyo falanqayn shaqaynaya.
Amaan: Bartosz Naskręcki(ku furmaa daaqad cusub)
Weydiin: # Isgoyska oogada algebraic geometry
Samee app sawira laba oogadood oo quadratic ah isla markaana casaanka ku midabeeya qalooca isgoyska. Adeegso computational Riemann-Roch theorem si tan loogu beddelo qalooc Weierstrass ah.
## Daaqadda weyn
Laba oogadood oo midab khafiif ah leh oo leh hadh yar oo hufan, sawirid tayo sare leh oo ku kulmaya qalooc algebraic oo casaan ah
Wareejin jiirka labada jiho, hab pinch buuxa oo zoom ah, taabasho haptic ah si loo muujiyo menu yar oo leh sliders lagu beddelayo coefficients-ka oogada kasta; ogaansho via heerka Z-buffor
## Daaqadda dhinaca midig
Xisaabta gaaban ee Weierstrass (oo ka sarreysa Q ama fidin field quadratic ah) oo si joogto ah loogu xisaabinayo iyada oo loo marayo formulas effective Riemann-Roch theorem
## Hab ambient ah halkaas oo dhammaan xakameeyayaashu qarsoon yihiin isticmaaluhuna uu ku raaxaysan karo quruxda qaababka
## Tilmaamo
App-ku wuxuu ku shaqaynayaa browser-ka, hirgelin fudud oo leh maktabadihii ugu dambeeyay ee full stack, la qaadi karo, la daabici karo
## Dukumiintiyo
Git repo, journal, plan (faylal Markdown)
“It’s incredibly energizing to use OpenAI’s new GPT-5.5 model in our harness, have it reason over massive biochemical datasets to predict human drug outcomes, and then see it deliver significant accuracy gains on our hardest drug discovery evals. If OpenAI keeps cooking like this, the foundations of drug discovery will change by the end of the year.”
U adeegista GPT‑5.5 iyadoo leh dib-u-dhaca GPT‑5.4 waxay u baahatay dib-u-fikir ku saabsan inference-ka sidii nidaam isku dhafan, ee ma aha qaybo hagaajinno gooni-gooni ah ah. GPT‑5.5 si wadajir ah ayaa loogu naqshadeeyay, loogu tababaray, looguna adeegay nidaamyada NVIDIA GB200 iyo GB300 NVL72. Codex iyo GPT‑5.5 waxay door muhiim ah ku lahaayeen sida aan u gaarnay yoolalkayaga waxqabad. Codex waxay kooxda ka caawisay inay si dhaqso leh uga gudbaan fikrad ilaa hirgelin la cabbiri karo, iyadoo sawiraysa habab, xididaysa tijaabooyin, kana caawinaysa aqoonsiga hagaajinada mudan maalgashi qoto dheer. GPT‑5.5 waxay ka caawisay helitaanka iyo hirgelinta horumarrada muhiimka ah ee stack-ga laftiisa. Si fudud, noocku wuxuu gacan ka geystay hagaajinta kaabeyaasha isaga adeega.
Mid ka mid ah horumarradaas wuxuu ahaa heuristics-ka isu-dheellitirka culeyska iyo kala-qaybinta. Ka hor GPT‑5.5, waxaan codsiyada ku kala qaybin jirnay accelerator-ka tiro go’an oo qaybo ah si shaqada loogu dheellitiro xudunta xisaabinta, anagoo xaqiijinayna in codsiyada waaweyn iyo kuwa yaryarba ku wada shaqayn karaan isla GPU-ga. Hase yeeshee, tiro hore loo sii go’aamiyay oo qaybo taagan ahi ma aha tan ugu habboon dhammaan qaababka taraafikada. Si GPU-yada si ka fiican looga faa’iidaysto, Codex waxay falanqaysay qaababka taraafikada wax-soo-saarka ee toddobaadyo badan waxayna qortay algorithmo heuristic gaar ah si shaqada si ugu habboon loo kala qaybiyo loona dheellitiro. Dadaalkani wuxuu yeeshay saameyn aad uga weyn intii la filayay, isagoo kordhiyay xawaaraha soo saarista tokennada in ka badan 20%.
Diyaarinta adduunka noocyo si aad u fiican ugu fiican helitaanka iyo xiritaanka nuglaanta amniga waa ciyaar kooxeed waxayna u baahan doontaa nidaamka oo dhan inuu si adag uga shaqeeyo dhisidda adkaysi, iyadoo la helayo gelitaan nooc oo la dimoqraadiyeeyay iyo hirgelin isdaba-joog ah oo loogu talagalay xilligga xiga ee difaaca cyber-ka.
Noocyada ugu casriyeysan waxay sii kordhinayaan awoodahooda amniga internetka. Awoodahaas si ballaaran ayaa loo qaybin doonaa, annaguna waxaan aaminsanahay in jidka ugu fiican ee hore loo maro uu yahay in la hubiyo in loo adeegsan karo dardargelinta difaaca cyber-ka iyo xoojinta nidaamka guud.
GPT‑5.5 waa tallaabo koror ah laakiin muhiim ah oo loo qaaday dhanka AI xallin karta qaar ka mid ah caqabadaha ugu adag dunida sida amniga internetka. Iyadoo GPT‑5.2 bishii Diseembar, waxaan si firfircoon u hirgelinnay ilaalinnada cyber-ka ee lagama maarmaanka u ah in lagu xaddido si-xun-u-isticmaalka cyber-ka ee noocyadeenna; hadda GPT‑5.5, waxaan hirgelinaynaa kala-soocayaal adag oo loogu talagalay khatarta cyber-ka ee suurtagalka ah kuwaas oo isticmaaleyaasha qaar marka hore dhib u arki karaan, inta aan waqti ka dib hagaajinayno.
Waxaan sannado badan u aqoonsanaynay amniga internetka inay tahay qayb ka mid ah Qaab-dhismeedka Diyaar-garowga(ku furmaa daaqad cusub) iyadoo noocyadeennu si tartiib tartiib ah u soo hagaageen, halka aan si isdaba-joog ah u horumarinaynay una hagaajinaynay yareynno, si aan mas’uuliyad leh ugu sii dayn karno noocyo leh awoodo macno leh oo amniga internetka ah.
- Waxaan hirgelinaynaa ilaalinno heer warshadeed hoggaaminaya oo ku habboon heerkan awoodda cyber-ka. Waxaan markii ugu horreysay soo bandhignay ilaalinno u gaar ah cyber-ka anagoo adeegsanayna GPT‑5.2(ku furmaa daaqad cusub) sanadkii hore, kuwaas oo aan sii wadnay tijaabintooda, hagaajintooda, iyo ku dhisiddooda hirgelinnadii xigay. GPT‑5.5, waxaan u qaabaynay xakamaynno ka adag hawlaha halista sare leh, codsiyada cyber-ka xasaasiga ah, waxaana ku darnay ilaalinno ka dhan ah si-xun-u-isticmaal soo noqnoqda. Helitaan ballaaran waxaa suuragelinaya maalgashigayaga badbaadada nooca, isticmaal la xaqiijiyay, iyo kormeerka isticmaalka aan la oggolayn. Waxaan bilo badan la shaqaynaynay khubaro dibadeed si loo horumariyo, loo tijaabiyo, loona ku celceliyo adkeysiga ilaalimahan. GPT‑5.5, waxaan hubinaynaa in horumariyayaashu si fudud u sugi karaan koodhkooda, iyadoo la saarayo xakamaynno adag qulqul-shaqooyinka cyber-ka ee ay u badan tahay inay dhibaato keenaan marka ay adeegsadaan dad xaasidnimo leh.
- Waxaan ballaarinaynaa helitaanka si loo dedejiyo difaaca cyber-ka heer kasta. Waxaan noocyadeenna u oggolaan badan cyber-ka ka dhigaynaa kuwo la heli karo iyada oo loo marayo Trusted Access for Cyber, annagoo ka bilaabayna Codex, taas oo ay ku jirto helitaan la ballaariyay oo loogu talagalay awoodaha amniga internetka ee horumarsan ee GPT‑5.5 oo leh xaddidaado yar oo loogu talagalay isticmaaleyaal la xaqiijiyay oo buuxiya qaar ka mid ah calaamadaha kalsoonida(ku furmaa daaqad cusub) marka la bilaabayo. Ururrada mas’uulka ka ah difaaca kaabeyaasha muhiimka ah waxay codsan karaan helitaanka noocyo u oggolaan badan cyber-ka sida GPT‑5.4‑Cyber, iyagoo buuxinaya shuruudo amni oo adag si ay noocyadan ugu isticmaalaan sugidda nidaamyadooda gudaha. Tani waxay siisaa difaacayaal badan oo la xaqiijiyay qalab awood badan oo loogu talagalay shaqo amni oo sharci ah iyadoo la yareynayo caqabado aan loo baahnayn si loo dimoqraadiyeeyo helitaanka awoodaha difaac ee muhiimka ah. Isticmaalayaashu waxay ka codsan karaan helitaan la aamini karo chatgpt.com/cyber(ku furmaa daaqad cusub) si loo yareeyo diidmooyinka aan loo baahnayn marka ay GPT‑5.5 u isticmaalayaan shaqo difaac oo la xaqiijiyay.
- Waxaan la shaqaynaynaa la-hawlgalayaal dowladeed si loo caawiyo ilaalinta kaabeyaasha muhiimka ah ee dadweynaha. Si wadajir ah, waxaan sahminaynaa sida AI horumarsan u taageeri karto shaqada difaaca ee mas’uuliyiinta la aamini karo ee masuulka ka ah nidaamyada ay dadku ku tiirsan yihiin, laga bilaabo nidaamyada dijitaalka ah ee suga xogta muhiimka ah ee canshuur-bixiyayaasha ilaa shabakadda korontada iyo sahayda biyaha ee bulshooyinka deegaanka.
Waxaan awoodaha bayoolaji/kiimiko iyo amniga internetka ee GPT‑5.5 u tixgelineynaa Heer Sare sida ku cad Qaab-dhismeedka Diyaar-garowga(ku furmaa daaqad cusub). Inkastoo GPT‑5.5 aysan gaarin heerka awood cyber oo Critical ah, qiimaynteena iyo tijaabadeennu waxay muujiyeen in awoodaheeda cyber-ku ay yihiin tallaabo koror ah marka loo eego GPT‑5.4.
Intaa waxaa dheer, GPT‑5.5 waxay martay geeddi-socodkeena buuxa ee badbaado iyo maamul ka hor sii-daynta, oo ay ku jiraan qiimayno diyaar-garow, tijaabinno gaar u ah domain-ka, qiimayno cusub oo bartilmaameed u ah awoodaha horumarsan ee bayoolajiga iyo amniga internetka, iyo tijaabo adag oo lala sameeyay khubaro dibadeed. Faahfaahin dheeraad ah waxaan ku wadaagnaa kaarka siistamka(ku furmaa daaqad cusub) ee GPT‑5.5.
Shaqadani waxay ka tarjumaysaa habkeena ballaaran ee adkeysiga AI, oo aan aaminsanahay in loo baahan yahay marka awoodaha noocyadu horumarayaan. Waxaan rabnaa in AI awood badan ay diyaar u noqoto dadka u adeegsanaya difaaca nidaamyada, hay’adaha, iyo dadweynaha. Jidka macquulka ahi waa helitaan la aamini karo, ilaalinno adag oo la jaanqaadaya awoodda, iyo awood hawlgal oo lagu ogaado laguna jawaabo si-xun-u-isticmaalka culus.
Maanta, GPT‑5.5 waxa loo soo bandhigayaa isticmaalayaasha Plus, Pro, Business, iyo Enterprise ee ChatGPT iyo Codex, halka GPT‑5.5 Pro loo soo bandhigayo isticmaalayaasha Pro, Business, iyo Enterprise ee ChatGPT. Waxaan GPT‑5.5 iyo GPT‑5.5 Pro u keenaynaa API-ga si aad ah dhow.
Gudaha ChatGPT, GPT‑5.5 Thinking waxa heli kara isticmaalayaasha Plus, Pro, Business, iyo Enterprise. GPT‑5.5 Pro, oo loogu talagalay su’aalo ka sii adag iyo shaqo saxnaan sare leh, waxa heli kara isticmaalayaasha Pro, Business, iyo Enterprise.
Gudaha Codex, GPT‑5.5 waxa loo heli karaa qorshayaasha Plus, Pro, Business, Enterprise, Edu, iyo Go iyadoo leh daaqad macne guud oo 400K ah. GPT‑5.5 sidoo kale waxa laga heli karaa habka Fast, iyadoo soo saaraysa tokenno 1.5x ka dheereeya qiime 2.5x ka badan.
Horumariyeyaasha API-ga, gpt-5.5 dhawaan waxa laga heli doonaa Responses API iyo API-ga dhammeystirka wada-sheekeysiga qiime dhan $5 halkii 1M token oo gelin ah iyo $30 halkii 1M token oo soo-saaris ah, iyadoo leh daaqad macne guud oo 1M ah. Qiimaha Batch iyo Flex waxa la heli karaa kala bar heerka caadiga ah ee API-ga, halka farsamaynta Priority laga heli karo 2.5x heerka caadiga ah. Waxaan sidoo kale sii dayn doonaa gpt-5.5-pro gudaha API-ga si loo helo saxnaan ka sii sarreysa, qiimihiisuna yahay $30 halkii 1M token oo gelin ah iyo $180 halkii 1M token oo soo-saaris ah. Faahfaahinta oo dhan ka eeg bogga qiimeynta.
Inkasta oo GPT‑5.5 laga qiimeeyo ka sarreeya GPT‑5.4, haddana waa mid ka caqli badan oo aad uga hufan token ahaan. Gudaha Codex, si taxaddar leh ayaanu u hagaajinay waayo-aragnimada si GPT‑5.5 ay ugu keento isticmaalayaasha intooda badan natiijooyin ka fiican iyadoo adeegsanaysa tokenno ka yar GPT‑5.4, isla markaana aan sii wadno bixinta isticmaal deeqsinimo leh oo heerarka is-diiwaangelinta oo dhan ah.
Kood qorid
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
SWE-Bench Pro (Dadweyne) * | 58.6% | 57.7% | - | - | 64.3% | 54.2% |
Terminal-Bench 2.0 | 82.7% | 75.1% | - | - | 69.4% | 68.5% |
Expert-SWE (Gudaha) | 73.1% | 68.5% | - | - | - | - |
*Shaybaarro ayaa ku xusay caddeyn xafidid(ku furmaa daaqad cusub) oo ku jirta qiimayntan
Xirfadeed
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
GDPval (guulo ama barbaro) | 84.9% | 83.0% | 82.3% | 82.0% | 80.3% | 67.3% |
FinanceAgent v1.1 | 60.0% | 56.0% | - | 61.5% | 64.4% | 59.7% |
Hawlaha Qaabaynta Bangiyada Maalgashiga (Gudaha) | 88.5% | 87.3% | 88.6% | 83.6% | - | - |
OfficeQA Pro | 54.1% | 53.2% | - | - | 43.6% | 18.1% |
Isticmaalka kombiyuutarka iyo aragga
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
OSWorld-Verified | 78.7% | 75.0% | - | - | 78.0% | - |
MMMU Pro (qalab la’aan) | 81.2% | 81.2% | - | - | - | 80.5% |
MMMU Pro (qalab leh) | 83.2% | 82.1% | - | - | - | - |
Isticmaalka qalabka
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
BrowseComp | 84.4% | 82.7% | 90.1% | 89.3% | 79.3% | 85.9% |
MCP Atlas** | 75.3% | 70.6% | - | - | 79.1% | 78.2% |
Toolathlon | 55.6% | 54.6% | - | - | - | 48.8% |
Tau2-bench Telecom*** | 98.0% | 92.8% | - | - | - | - |
** MCP Atlas: natiijooyinka Scale AI kadib cusboonaysiintii ugu dambeysay ee Abriil 2026.
*** Tau2-bench telecom: natiijooyinka 5.5 iyo 5.4 oo leh weydiimihii asalka ahaa, taas oo ah iyada oo aan wax hagaajin weydiin ah la samayn. Tani waxay ka reebaysaa natiijooyinka shaybaarro kale oo lagu qiimeeyay hagaajin weydiin ah.
Tacliimeed
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
GeneBench | 25.0% | 19.0% | 33.2% | 25.6% | - | - |
FrontierMath Tier 1–3 | 51.7% | 47.6% | 52.4% | 50.0% | 43.8% | 36.9% |
FrontierMath Tier 4 | 35.4% | 27.1% | 39.6% | 38.0% | 22.9% | 16.7% |
BixBench | 80.5% | 74.0% | - | - | - | - |
GPQA Diamond | 93.6% | 92.8% | - | 94.4% | 94.2% | 94.3% |
Humanity's Last Exam (qalab la’aan) | 41.4% | 39.8% | 43.1% | 42.7% | 46.9% | 44.4% |
Humanity's Last Exam (qalab leh) | 52.2% | 52.1% | 57.2% | 58.7% | 54.7% | 51.4% |
Amniga internetka
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
Hawlaha caqabadda Capture-the-Flags (Gudaha)**** | 88.1% | 83.7% | - | - | - | - |
CyberGym | 81.8% | 79.0% | - | - | 73.1% | - |
**** Ballaarin ka mid ah CTF-yadii ugu adkaa ee lagu isticmaalay kaararka siistamka oo leh caqabado adag oo dheeraad ah.
Macnaha dheer
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
Graphwalks BFS 256k f1 | 73.7% | 62.5% | - | - | 76.9% | - |
Graphwalks BFS 1mil f1 | 45.4% | 9.4% | - | - | 41.2% (Opus 4.6) | - |
Graphwalks parents 256k f1 | 90.1% | 82.8% | - | - | 93.6% | - |
Graphwalks parents 1mil f1 | 58.5% | 44.4% | - | - | 72.0% (Opus 4.6) | - |
OpenAI MRCR v2 8-needle 4K-8K | 98.1% | 97.3% | - | - | - | - |
OpenAI MRCR v2 8-needle 8K-16K | 93.0% | 91.4% | - | - | - | - |
OpenAI MRCR v2 8-needle 16K-32K | 96.5% | 97.2% | - | - | - | - |
OpenAI MRCR v2 8-needle 32K-64K | 90.0% | 90.5% | - | - | - | - |
OpenAI MRCR v2 8-needle 64K-128K | 83.1% | 86.0% | - | - | - | - |
OpenAI MRCR v2 8-needle 128K-256K | 87.5% | 79.3% | - | - | 59.2% | - |
OpenAI MRCR v2 8-needle 256K-512K | 81.5% | 57.5% | - | - | - | - |
OpenAI MRCR v2 8-needle 512K-1M | 74.0% | 36.6% | - | - | 32.2% | - |
Caqliyeyn aan la taaban karin
Qiimayn | GPT‑5.5 | GPT‑5.4 | GPT‑5.5 Pro | GPT‑5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro |
ARC-AGI-1 (La xaqiijiyey) | 95.0% | 93.7% | - | 94.5% | 93.5% | 98.0% |
ARC-AGI-2 (La xaqiijiyey) | 85.0% | 73.3% | - | 83.3% | 75.8% | 77.1% |
Qiimaynta GPT waxaa lagu waday effort-ka caqliyeynta oo lagu dejiyay xhigh, waxaana lagu qabtay jawi cilmi-baariseed, taas oo mararka qaar bixin karta wax-soo-saar wax yar uga duwan kan ChatGPT wax-soo-saarka.
