GPT-6 Astra: A new generation of intelligence
Jiil cusub oo Garaadka ah
Waxaan soo bandhigaynaa GPT‑6 Astra, oo ah moodeelka ugu caqliga badan uguna iswaafaqsan adduunka.
GPT‑6 Astra waxay isu keentaa sannado cilmi-baaris ah iyo maalgashiyo waaweyn oo ku saabsan wejiga hore ee tabbabarka, waxbarashadda xoojinta ah, iyo waafajinta. Astra waxay ku jirtaa heerka ugu horumarsan ee adeegsiga kombiyuutarka, daalacashada internetka, injineernimada softiweerka, amniga internetka, sayniska, iyo shaqada xirfadeed. Astra waxay FrontierMath Tier 4 ka heshay dhibco 98%, waxayna horeba gacan uga gaysatay xallinta mas'alooyin furan oo muddo dheer jiray oo ka tirsan cilmiga xisaabta. Astra waxay sidoo kale ARC-AGI-3 ka gaadhay dhibco 99.9% ah, ExploitBench-na dhibco 100% ah. Waxay sidoo kale dejisaa ugu casriyeysan oo adeegsiga kombiyuutarka iyo biraawsarka, iyadoo xawaare, saxnaan, iyo go'aan-qaadasho aan lala sinnayn ku maareynaysa shaqada xirfadeed ee ugu baahida badan.
GPT‑6 Astra maanta si tartiib-tartiib ah ayaa loogu sii daynayaa tiro xaddidan oo ururro ah, maalmaha soo socdana waxa heli doona dhammaan isticmaalayaasha ChatGPT Plus, Pro, Business, iyo Enterprise, iyo sidoo kale iyada oo loo marayo OpenAI API, Microsoft Azure, iyo AWS Bedrock.
Astra waa moodalladeena ugu jaangoan, iyadoo horumar weyn laga sameeyey fahamka ujeeddada isticmaalaha iyo hab-dhaqanka moodalka—waxaad hawlaha ugu wakiilan kartaa adigoo kalsooni weyn ku qaba go'aanka Astra. Hal dariiqo oo aan tan ku tijaabinno, waxaan dhisnay qiimayn cusub oo laga duulay dhacdadii Hugging Face, taas oo qiimaynaysa in moodal wajahaya hawl adag ama aan suurtagal ahayn uu ka gudbi doono xadka loogu talagalay. Marka la barbar dhigo GPT‑5.6 Sol, oo iyada oo aan lahayn ilaalinta wax-soo-saarka ka gudubtay bartilmaameedka la oggolaaday 48% kiisaska, GPT‑6 Astra waxay tan samaysay 0% kiisaska.
Nooc ugu fiican adduunka ee isticmaalka kombiyuutarka
GPT‑6 Astra waxay astaan u tahay horumar cusub oo ugu casriyeysan xawaaraha, saxnaanta, iyo badbaadada isticmaalka kombiyuutarka. Waxay qaban kartaa hawlo daal badan sida buuxinta foomamka onlaynka ah, cusboonaysiinta diiwaannada macaamiisha ee CRM-ka, iyo habaynta jadwalkaaga. Waxay sameyn kartaa cilmi-baaris khadka internetka ah, waxayna ku diyaarin kartaa soo koobidyo iimaylkaaga ama tifaftiraha dukumeentigaaga. Waxay falanqayn kartaa xog cilmiyeed, soo saari kartaa garaafyo, samayn kartaa mareeg, isla markaana fulin kartaa hubinnada QA ee shaashadda hore si loo xaqiijiyo in dhammaan astaamaha mareegtaas ay shaqaynayaan. Waxay kaa caawin kartaa inaad si madax-bannaan u rakibto oo u tijaabiso softiweerka, iyo inaad xalliso dhibaatooyinka aad shaashadda ku aragto. Horumaradan waxay sidoo kale ka muuqdaan natiijooyinkeenna qiimaynta ee heerka ugu casrisan.
Hagaajinnadani waxay sidoo kale keenaan koror weyn oo wax-ku-oolnimo ah hawlaha dhabta ah ee shaqada aqooneed. Jilitaannada waqtiga jawaabta ee OSWorld 2.0, Astra wuxuu ku gaaraa waxqabad sare oo adeegsiga kombiyuutarka ah qiyaastii 47% waqti ka yar hawl kasta marka loo eego GPT‑5.6 Sol, oo gaartay 72.6% iyadoo hawl kasta qaadanaysa qiyaastii 40 daqiiqo, marka la barbar dhigo 65.7% iyadoo qaadanaysa qiyaastii 75 daqiiqo.3
Awoodaha GPT‑6 Astra ee isticmaalka kombiyuutarka waxaa laga arki karaa natiijooyinka qaybaha kala duwan, oo ay ku jiraan horumarinta ciyaaraha, injineernimada korontada, iyo shaqada aqoonta ee maalinlaha ah:
Alongside Astra, we are also updating the Codex harness to significantly improve the speed of computer use. Combined with Astra’s efficiency, this translates to a 1.9x faster task completion compared to the current GPT‑5.6 Sol experience, on the Mind2Web benchmark. The model’s improvements on speed mean it can take on many time-consuming life tasks for you, faster than you can.4
Isbeddel weyn oo ku yimid shaqada xirfadeed
GPT‑6 Astra waxa uu isku daraa horumarrada adeegsiga kombiyuutarka iyo tababar bartilmaameed leh oo loogu talagalay deegaannada xirfadeed, si uu uga caawiyo wax ka qabashada hawlaha shaqo ee adag. Waxa uu isku daraa garaadka looga baahan yahay dhibaatooyinka adag iyo awoodda lagu fulin karo habraacyo shaqo oo tallaabooyin badan leh, laguna soo saari karo dukumentiyo, xaashiyo xisaabeed, iyo bandhigyo tayo sare leh.
GPT‑6 Astra waa moodelkeenna ugu fiican ee u hoggaansanaanta qaab-dhismeedyada jira iyo soo saarista islaydho si wanaagsan loo habeeyey oo si kooban u gudbiya qodobbada muhiimka ah iyada oo loo marayo sheeko habaysan. Waxay abuurtaa dukumentiyo cad oo si wanaagsan loo habeeyey, bandhigyo, xaashiyo xisaabeed, iyo falanqayno raaca qaab-dhismeedyadaada oo la jaanqaada hab-qoraalkaaga iyo qaabkaaga muuqaalka. Astra sidoo kale waxaa loo tababaray inay si gaar ah natiijooyinka ugu soo darto oo keliya macnaha guud ee muhiimka ah, halkii ay ku celcelin lahayd macluumaad aan loo baahnayn hawsha markaas la qabanayo. Dhammaan tani waxay ka dhigan tahay inay soo saari karto agabyo isla markiiba la isticmaali karo oo waafaqsan macnaha ganacsigaaga iyo heerarkaaga.
GPT‑6 Astra sidoo kale waxay u keentaa qiimeyn muuqaal oo ka xoog badan mareegaha, ciyaaraha, ababka, iyo muuqaal-soo-bandhigyada ay dhisto. Gudaha ChatGPT, Astra waxay adeegsan kartaa Sites(ku furmaa daaqad cusub) si ay si toos ah uga abuurto weydiin mareegaha, ababka webka, iyo ciyaaraha, u martigeliso, una wadaagto.
Marka tilmaamuhu ay bannaanka u reebaan fasiraad, GPT‑6 Astra ayaa moodalladii hore uga fiican samaynta go'aanka saxda ah. Waxay isticmaashaa macnaha guud si ay u buuxiso meelaha caadiga ah ee ka dhiman, waxayna weydiisaa su’aalo si gaar ah u diiradda saaraya marka jawaabtu ay beddeli karto natiijada. Codex gudaheeda, waxay su'aalo u weydiin kartaa si aan isku mar ahayn iyadoo sii wadaysa shaqada aan ku xirnayn jawaabtaada. Haddii aadan jawaabin, nidaamku wuu sii socdaa isagoo sameynaya mala-awaallo macquul ah halka ay ku habboon tahay, balse wuxuu sugayaa talo-gelintaada marka ay timaaddo go’aannada saamaynta weyn leh.
Tusaalooyinka hoose waxay muujinayaan sida Astra uga wada shaqayso hawlaha maalinlaha ah oo xog maqan ay si la taaban karo u beddeli karto jawaabta.
Astra sidoo kale way ka wanaagsan tahay inay sii ilaaliso jihada marka hawshu is beddesho. Moodalladii hore mararka qaarkood waxay farriimaha hagidda ula dhaqmi jireen yool cusub, iyagoo luminaya raadkii codsigii asalka ahaa ama xaddidaadihii hore. Astra waxa uu ku daraa shuruudaha cusub, waxa uu beddelaa jihada marka la weydiisto, wuxuuna ka jawaabaa su’aalaha dhinaca ah isaga oo aan ka tagin hawsha guud.
Kood qorista
GPT‑6 Astra waa nooca ugu fiican ee injineernimada softiweerka ilaa hadda.
“GPT‑6 Astra wuxuu bixiyaa waxqabad heerkii ugu sarreeyay ee tijaabooyinkayaga gudaha ee koodhaynta, wuxuuna muujinayaa horumar cad oo ku yimid qiimeynta dareenka ganacsiga marka loo eego GPT‑5.6 Sol. Marka loo adeegsado koodhaynta wakiil-ku-saleysan, GPT‑6 Astra wuxuu u wada xiriiraa hab ay horumariyayaashu si fudud ula socon karaan, wuxuuna soo saaraa koodh u baahan ku-celcelin yar si uu u gaaro tayada wax-soo-saarka.”
“Astra waxaan ku tijaabinnay heerarka dadaalka hoose, dhexdhexaad, iyo sare mid ka mid ah qiimeynahayagii jiilka koowaad, waxayna si weyn uga sarreysay GPT 5.6 Sol. Dadaalka sare wuxuu keenaa wareegyo dheeraad ah oo lagu hagaajiyo dhisme cusub, xaqiijin dheeraad ah oo lagu sameeyo tijaabinta browser-ka, iyo u janjeerid dhanka fulinta koodhka halkii laga isticmaali lahaa apply-patch. Fahamka sida nooc u isticmaalo dadaalkiisa ayaa noo suurtageliya inaan malaayiin dhisayaal ah siino waddo ka degdeg badan oo la isku halayn karo oo ay uga gudbaan fikrad una gaaraan app shaqaynaya.”
Iyadoo la adeegsanayo Astra, waxaan soo bandhigaynaa hab cusub oo Codex uu ku kaydin karo uguna soo ceshan karo macnaha guud marka shaashada macnaha guud buuxsanto. Taariikh ahaan, moodalladu waxay adeegsadeen isku-soo-ururin si ay shaqada u soo koobaan inta lagu jiro kalfadhiyada dhaadheer, sida marka la cilad-bixinayo dhibaatooyinka murugsan ama la qabanayo dib-u-habeynno waaweyn. Isku-soo-ururin kastaa waxay ka tagi kartaa faahfaahin ku saabsan sababta saxitaanku u fashilmay ama sida ay qayb u dhaqanto. Codex gudaheeda, Astra waxay ku hayn kartaa qoraallo shaashada macnaha guud, iyadoo ilaalinaysa faahfaahinta ururtay iyada oo aan marar badan lagu cadaadin hal soo koobid. Shaashada macnaha guud hore weli waa la raadin karaa, sidaas darteed Astra waxay ka heli kartaa shuruudaha ama natiijooyinka tijaabada farriimihii hore iyo wax-soo-saarka qalabka—xitaa haddii macluumaadkaas aan lagu darin qoraalladeeda. Waxaad awoodsiin kartaa sifadan tijaabada ah ee ku jirta config.toml-ka Codex,(ku furmaa daaqad cusub) waxayna noqon doontaa midda caadiga ah ee Astra toddobaadyada soo socda.
Horumarinta daahfurka sayniska
Astra waxay ka caawin kartaa hawlaha wax-ku-oolka ah ee ka dambeeya daahfurka sayniska. Iyadoo isku daraysa caqliyeynta sayniska iyo isticmaalka kombiyuutarka, waxay si toos ah uga shaqayn kartaa software khaas ah si ay u baadho xogta una sahmiso natiijooyinka, iyadoo ka caawinaysa cilmi-baarayaasha inay qiimeeyaan caddaymaha oo go’aamiyaan waxa xiga ee la baarayo.
Amniga internetka
As we discussed in our safety update, Astra is a significant jump in cyber capabilities and meets the Critical threshold in cybersecurity under our Preparedness Framework. Its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but it also creates a need for stronger safeguards. To understand how far these capabilities extend, we ran Astra on internal and third-party expert evaluations.
We first tested the model without production safeguards on ExploitBench and ExploitGym, which evaluate whether models can turn known software vulnerabilities into working exploits. On ExploitBench, Astra achieved a perfect score of 100%, compared with 78.5% for GPT‑5.6 Sol, our previous frontier cyber-capable model. On ExploitGym, Astra reached a 42.4% success rate, compared with 30.3% for GPT‑5.6 Sol, while using substantially fewer output tokens.13
Given concerns that exposure to historical software vulnerabilities may have affected benchmark results, we also evaluated Astra on two novel benchmarks. For one, we built an internal “ExploitBench (June–August 2026)” evaluation to test exploit development using vulnerabilities from the previous three months.14 Astra achieved substantially higher arbitrary code-execution rates than GPT‑5.6 Sol on this dataset while using far fewer output tokens. During the evaluation, Astra even discovered and used two previously unknown zero-day vulnerabilities. We are disclosing both vulnerabilities to their maintainers.
We also tested Astra on SRE-Bench15, a benchmark that measures whether models can reverse engineer software binaries to understand its core logic without access to raw source code. Astra solved 88.0% of tasks in a single attempt and 99.2% within four attempts, compared with 55.9% and 68.7% for GPT‑5.6 Sol, respectively.
Beyond benchmarks, expert-led assessments found that Astra, when run without production safeguards, could use previously unknown vulnerabilities to achieve arbitrary code execution in hardened browsers and create privilege-escalation exploits for hardened operating-systems.
As we discussed in The Defender’s Window, frontier cyber capabilities can help defenders find weaknesses faster, but they also make those weaknesses easier to exploit, raising the urgency for defenders to adapt. With the version of Astra launching today, defenders can use it to complete tasks such as secure code review and patching.
However, Astra will refuse to comply with more advanced cybersecurity tasks such as creating proof-of-concept exploits for vulnerabilities. Through OpenAI Daybreak, we plan to expand access and roll out less restrictive safeguards in the coming weeks. This will enable more defensive workflows, including vulnerability and proof-of-concept validation, malware analysis, and detection engineering.
We have also strengthened our protections against potential cyber misuse, building upon our safeguards stack for GPT‑5.6 Sol. These include stronger model robustness to better withstand potential jailbreaks and more context for our monitoring systems. We have continued rigorous internal and external testing, including automated evaluations with our internal red-teaming attackers. More details about our cyber safeguards and testing are available in the Astra safety overview and system card(ku furmaa daaqad cusub)..
Iswaafajinta iyo si masuuliyad leh u daah furidda GPT‑6 Astra
Astra waa moodeelkayaga ugu waafaqsan. Astra waxay aad ugu fiican tahay inay taxaddar muujiso, ixtiraamto xuduudaha hawsha, isla markaana si daahfuran u wada xiriirto. Shaqadani waa wax-soo-saarkii ugu dambeeyay ee barnaamijkeenna cilmi-baarista ee muddada dheer socday, kaas oo diiradda saaraya tababarka moodeellada ku sii waafaqsan ujeeddada aadanaha min bilow ilaa dhammaad.
Deegaannada xasaasiga ah, Astra waxay ku dhaqaaqdaa taxaddar u dhigma heerka khatarta ay leedahay. Qiimayn lagu sameeyay hawlo isticmaalka kombiyuutarka ah oo si liddi ku ah loogu soo xulay inay kiciyaan habdhaqan khaldan, Astra waxay si ka guul badan uga fogaatay cawaaqib aan loogu talagalin. Ku shaqaynta tallaabooyin amni oo dheeraad ah oo si caadi ah loo bixiyo waxay keentay waxqabad xitaa ka xoog badan.
Astra causes fewer misaligned outcomes than any other frontier models tested. For a fair comparison, we used a generic computer-using-agent harness (based on the native computer use tools available in both the OpenAI Responses API(ku furmaa daaqad cusub) and Anthropic Messages API(ku furmaa daaqad cusub)) and without additional protections that are normally deployed to Codex and ChatGPT Work users (auto-review, confirmation policy). 16
Astra sidoo kale waxay u badan tahay inay ku shaqeyso gudaha xuduudaha uu dejiyay isticmaaluhu iyo kuwa ay tilmaamayso deegaankeeda. Qiimeyn gudaha ah, Astra marna iskuma dayin inay ka gudubto diidmada Codex Auto-Review. Tani way sii jirtay xitaa markii Auto-review si ula kac ah loogu habeeyey in laga leexan karo, hawshuna aanay suurtagal ahayn in si kale loo dhammaystiro. Ixtiraamkan ay u hayso xannibaadaha deegaanka wuxuu waafaqsan yahay natiijooyinka qiimeyntayada hawlaha sayberka ee aan suurtagalka ahayn, oo aan ku wadaagnay hordhaca qoraalkan, iyo natiijooyin kale oo lagu diiwaangeliyey kaarka siistamka(ku furmaa daaqad cusub).
Astra additionally sets new marks in transparent user communication. In one evaluation, Astra is three times less likely than GPT‑5.6 Sol to make inaccurate representations about its capabilities and affordances.
Qiimeyntayada dhalanteedka awoodaha, Astra wuxuu muujinayaa horumar la taaban karo marka loo eego GPT‑5.6 Sol, isagoo samaynaya sheegashooyin marin habaabin ah oo yar oo ku saabsan awoodihiisa.
Our evaluations found Astra’s written reasoning harder to monitor than GPT‑5.6 Sol’s, based on tests that explicitly asked it to evade monitoring. We attribute this to Astra’s greater control over written reasoning on simpler tasks and ability to solve problems with fewer written steps. Astra still appears to struggle to conceal the reasoning needed for complex tasks, but we take the decline seriously. Improving monitorability remains a research priority, and the accompanying system card(ku furmaa daaqad cusub) details our findings and ongoing work.
Alignment training is core to our approach to deployment. As an additional layer of defenses, we also build system safeguards like Codex Auto-review(ku furmaa daaqad cusub) and monitoring agents’ reasoning and actions to help detect and contain unsafe behavior. As described in our safety update, we are also deploying misalignment monitoring in production for Astra-class models in order to have visibility into misalignment, and help contain its worst instances. These safeguards resemble our monitoring for internal deployments and involve a system of classifiers which check the model’s reasoning and actions for unauthorized behavior and automatically stop potentially unauthorized activity.
Given the significant increase in Astra’s cybersecurity capabilities, we are being especially careful to make this deployment safe and secure. Extra safety checks can sometimes slow, pause, or stop legitimate work, including defensive cybersecurity. If a task is paused in ChatGPT or Codex, you may be asked to review the action before continuing. In the API, the task will stop. These checks can sometimes interrupt legitimate work, and we are continuing to iterate on this system to reduce unnecessary interruptions. Misalignment monitoring cannot replace alignment: our goal is to build models that reliably stay within their authorized scope, so these protections do not need to intervene.
Helitaan
GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock. Astra usage is included within the existing subscription allowances—users and businesses will also be able to purchase credits for additional usage. Users on the Pro, Business, and Enterprise plans will also get access to GPT‑6 Astra Pro. Enterprise administrators can enable Astra for their workspace; access is off by default at launch.
Astra supports Zero Data Retention for eligible API customers, and as we shared last month, we're testing Private Safety Processing to strengthen safety monitoring while preserving customer privacy.
For developers, GPT‑6 Astra will be available in the OpenAI API as gpt-6-astra and through Microsoft Azure and Amazon Bedrock.
OpenAI API Standard pricing is $10 per million input tokens and $50 per million output tokens. Separate rates apply to cache reads and writes. Fast mode is available for GPT‑6 Astra in the API and delivers up to 2x the speed of Standard processing at 2x the Standard price.
Professional
Professional | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
AutomationBench | 41.4% | 18.1% | 31.4% | 17.4% | 26.9% | - |
BenchCAD | 95.9% | 83.3% | 84.3% 5 | 67.5% 5 | 82.1% 5 | - |
BrowseComp | 91.5% | 90.4% | - | 87.4% | 90.8% | - |
OpenScore String Quartets (1 - OMR-NED) | 0.84 | 0.19 | - | - | - | - |
Internal Design Tasks | 50.0% | 47.4% | - | 35.8% | - | - |
Internal Data Science Tasks | 40.9% | 30.5% | - | 34.7% | - | - |
Artificial Analysis Intelligence Index v4.1.1 | 61.2 | 60.9 | 65.7 | 62.1 | 63.1 | 58.7 |
Coding
| Coding | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
| Terminal-Bench 4.0 | 57.9% | 37.3% | 55.8% | 44.5% | 52.6% | 19.1% |
| DeepSWE v1.1 | 74.1% | 72.7% | 67.4% | 69.9% | 73.7% | 73.8% |
| FrontierCode 1.1 Extended (score) | 64.5% 8 | 60.6% | 63.6% | 64.9% | 63.6% | 56.3% |
| FrontierCode 1.1 Main (score) | 53.3% 8 | 47.5% | 50.9% | 53.5% | 53.4% | 43.6% |
| Internal Database Migration Tasks | 63.9% | 42.7% | 57.8% | 50.3% | - | - |
| Artificial Analysis Coding Agent Index v1.4 | 67.0 | 65.1 | - | 67.2 | 68.1 | 61.2 |
Academic
Academic | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
Terminal-Bench Science 0.1 | 64.6% | 22.4% | 52.6% | 21.4% | 30.0% | - |
FrontierMath Tier 4 (v2) | 97.6% | 83.0% | 87.8% | 90.2% | 73.2% | - |
GPQA Diamond | 96.0% | 94.6% | 93.7% | 92.6% | 93.7% | 95.3% |
Humanity's Last Exam (w/ tools) | 57.2% | - | 65.0% | 63.8% | 63.6% | - |
Habka
Habka | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
Halbeegga badqabka isticmaalka kombiyuutarka gudaha (inta uu hooseeyo ayaa fiican) | 2.4% | 22.0% | 9.5% | 18.3% | 11.5% | - |
Halbeegga badbaadada ee isticmaalka kombiyuutarka gudaha, oo leh AutoReview (hoose ayaa ka fiican) | 1.8% | 4.3% | - | - | - | - |
Halbeegga gudaha ee ka-wareegista (heerka hoose ayaa ka fiican) | 0.00% | 0.29% | - | - | - | - |
Nidaam-dabinta amniyeed ee ExploitGym (heerka hoose ayaa ka fiican) | 0.0% | 48.2% | - | - | - | - |
ExploitGym Aan Suurtagal Ahayn | 100.0% | - | - | - | - | - |
Halbeegga gudaha ee dhalanteedka (heerka hoose ayaa wanaagsan) | 4.2% | 12.2% | - | - | - | - |
Macnaha guud ee dheer
Macnaha guud ee dheer | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
OpenAI MRCR v2 8-needle 256K-512K | 100.0% | 91.5% | - | - | - | - |
OpenAI MRCR v2 8-needle 512K-1M | 96.3% | 73.8% | - | - | - | - |
Abstract reasoning
| Abstract reasoning | GPT‑6 Astra | GPT‑5.6 Sol | Claude Fable 5.1 | Claude Fable 5 | Claude Opus 5 | Gemini 3.8 Flash |
| ARC-AGI-3 | 99.9% 1 | 7.8% | - | - | 30.2% | - |
| ARC-AGI-2 | 95.0% | 92.5% | 90.0% | 89.2% | 90.4% | - |
| ARC-AGI-1 | 98.5% | 97.5% | 97.5% | 98.5% | 97.5% | - |
Dhibcaha qiimayntu waa kuwa ugu sarreeya heer kasta oo dadaal ah. Qiimaynnada GPT waxaa lagu sameeyay deegaankayaga cilmi-baarista ama iyada oo loo marayo API-gayaga, kuwaas oo laga yaabo inay bixiyaan wax-soo-saar wax yar ka duwan kan ChatGPT ee wax-soo-saarka, sababo la xiriira kala duwanaanshaha weydiimaha nidaamka, qalabka la heli karo, iwm.
FOOTNOTES
- 1
On ARC-AGI-3, GPT-6 Astra was run with our responses API harness, which changes two settings to better match real-world performance. The changes do not specifically target ARC-AGI-3.
- 2
GPT-5.6 Sol refers to the version available in our API, ChatGPT Codex, and ChatGPT Work. The version in ChatGPT Chat is slightly different.
- 3
OSWorld V2-Offline is a subset of the original OSWorld V2 that works without internet access. Claude model performance on OSWorld-V2 Offline was reproduced by the authors on the official leaderboard(ku furmaa daaqad cusub). On OSWorld 2.0, the scores for Claude use the official settings, and not the modified tasks and modified grading from the Fable 5.1 System Card.
- 4
- 5
On BenchCAD, Claude's scores reflect 3 modifications to the eval, detailed in the Fable 5.1 System Card(ku furmaa daaqad cusub).
- 6
Guang Yang, Victoria Ebert, Nazif Tamer, Brian Siyuan Zheng, Luiza Pozzobon, and Noah A. Smith. “LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR(ku furmaa daaqad cusub).” arXiv:2506.19065, 2025.
- 7
Mark R. H. Gotham, Maureen Redbond, Bruno Bower, and Peter Jonas. “The OpenScore String Quartet Corpus(ku furmaa daaqad cusub).” Proceedings of the 10th International Conference on Digital Libraries for Musicology, pp. 49–57. ACM, 2023.
- 8
On FrontierCode, GPT-6 Astra was run with a developer message similar to a section of its developer message in Codex(ku furmaa daaqad cusub): "Avoid creating excessive test files. Create a new test file only when required by repository conventions or when no existing file is a suitable home. Avoid unrelated cleanup and unnecessary complexity. Reuse suitable existing utilities. Read relevant repository instructions and inspect nearby code, tests, documentation, and CI. Follow established conventions. The goal is clean, mergeable code." The prompt was not optimized for the eval.
- 9
The first concerns how close together prime numbers can occur, however far along the number line you go. For more than a decade, the best known result established that infinitely many pairs of primes are at most 246 apart. Julia Stadlmann(ku furmaa daaqad cusub) recently improved that bound to 240. Astra helped establish a stronger bound of 186, showing that infinitely many pairs occur within this smaller distance. Short prime gaps: Proof(ku furmaa daaqad cusub) and supporting research(ku furmaa daaqad cusub).
- 10
The second concerns unusually large gaps between primes. Astra improved a term in a bound on these gaps that had remained unchanged for more than 80 years. We’re sharing the proofs and abridged chain of thought and verification materials for both results. Large prime gaps: Proof(ku furmaa daaqad cusub) and supporting research(ku furmaa daaqad cusub).
- 11
- 12
- 13
- 14
ExploitBench (June–August 2026) contains 20 high-severity V8 vulnerabilities across 13 stable Chrome releases. The benchmark tests whether agents can achieve arbitrary code execution in V8 and official Chrome releases for Linux by exploiting each specified vulnerability. Some included vulnerabilities may not permit arbitrary code execution under the evaluation’s constraints, so a 100% success rate may not be achievable. Note: the 5.5% score of GPT-5.6 Sol is an artifact of the 300-turn limit in the benchmark, which is not a limit that real customers using max would have. The model at similar settings achieved an 11.5% score when hitting fewer limits.
- 15
Jeremy Spence et al. “The Next Challenge for Agentic Cybersecurity: A Realistic, Contamination-Free Reverse Engineering Benchmark(ku furmaa daaqad cusub).” arXiv:2608.11469v1, 2026.
- 16
When we test across third-party models, we use a simpler research setup. Codex has a more complex production configuration, which can result in different raw-model error rates. Provider-side safeguards and computer-tool implementations still differ. Users do not experience the no-confirmation scenario in Codex, as it's an internal research configuration.
- 17
