We're beginning a limited preview of the GPT‑5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model. Terra has competitive performance to GPT‑5.5 while being 2x cheaper and Luna brings strong capability at our lowest cost.
GPT‑5.6 Sol launches with our most robust safety stack to date. We strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse, and spent multiple weeks finding weaknesses, pressure-testing our system, and hardening it against real-world attacks.
We believe in broad access, and we plan to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks. As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch. At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly. During this preview, we will continue testing and coordinating closely with partners as we work toward broader availability. We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.
GPT‑5.6 Sol is our strongest model yet. To give a preview of model performance, we share a set of evaluations highlighting improved agentic capabilities in coding, biology, and cybersecurity, with additional safety and preparedness evaluations available in our system card(otvara se u novom prozoru). We will share an expanded suite of evaluation results when we make the model broadly available.
With GPT‑5.6, we’re introducing a new max reasoning effort to give Sol the most time to reason deeply. Additionally, we’re introducing a new ultra mode that goes beyond the capabilities of a single agent by leveraging subagents to accelerate complex work.
For coding workflows, GPT‑5.6 Sol sets a new state of the art on Terminal‑Bench 2.1, which tests command-line workflows requiring planning, iteration, and tool coordination.
GPT‑5.6 Sol also shows broad improvements in biology workflows. On GeneBench v1, which evaluates long-horizon genomics and quantitative-biology analyses, it achieves stronger results than GPT‑5.5 while using fewer tokens.
GPT‑5.6 Sol is our most capable model yet for cybersecurity. It shifts the performance-efficiency frontier for long-horizon security tasks including vulnerability research and exploitation. On ExploitBench², GPT‑5.6 Sol is competitive with Mythos Preview using only ~1/3 of the output tokens. On ExploitGym(otvara se u novom prozoru)3, a benchmark created by UC Berkeley researchers in collaboration with OpenAI and other frontier labs, GPT‑5.6 Sol, Terra, and Luna models all demonstrate strong improvements in cyber capabilities as we increase reasoning.
We developed GPT‑5.6 Sol, Terra and Luna with our most robust safeguards to date, with configurations matched to each model’s capabilities. As the model becomes more capable, we design safeguards to increasingly hold up to real-world adversarial pressure while preserving access to legitimate work such as code review, vulnerability research, patch development, debugging, security education, and defensive testing. Our goal is to make prohibited offensive activity more difficult, uncertain, and detectable without unnecessarily limiting those beneficial uses. Based on our assessment of the model and safeguards, we expect substantial benefit for legitimate defensive work, while meaningfully constraining prohibited offensive use.
GPT‑5.6 Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks. As these capabilities continue to advance, our priority is to make sure they reach and benefit defenders, who can use these tools to find weaknesses, develop patches, and strengthen systems more broadly.
GPT‑5.6 Sol does not cross the Cyber Critical threshold under our Preparedness Framework. In evaluations involving Chromium and Firefox, it identified bugs and exploitation primitives—the building blocks of an exploit—but did not autonomously produce a functional full-chain exploit under the conditions tested. Still, benchmark thresholds cannot capture every way a model may be used or combined with other tools. That uncertainty, along with the model’s broader step change in capabilities, is why we are pairing the model’s increased capabilities with stronger safeguards and a phased release. We share more details about our safeguards in the GPT‑5.6 Preview system card(otvara se u novom prozoru).
No single safeguard is sufficient against determined or adaptive misuse. Across the GPT‑5.6 preview, we use layered safeguards, with exact configurations varying across models, and pressure-test them for real-world attacks. These include protections trained into the model, real-time checks during generation, account-level signals, differentiated access, monitoring, enforcement, and continued testing.
GPT‑5.6 is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model. These model-level safeguards establish the first boundary around what the model should and should not help with.
Real-time cyber and biology misuse classifiers provide another layer by evaluating output as it is generated. For higher risk cases, if they detect a potential violation, the generation may be paused while a larger reasoning model reviews the conversation and its context. If the output is assessed as disallowed, it is withheld before it reaches the user.
Flagged activity can also trigger account-level review across relevant conversations and risk signals, consistent with our terms and policies around content retention and review. Looking beyond a single conversation helps our systems distinguish persistent malicious behavior from legitimate dual-use security work, where similar technical concepts may appear in very different contexts.
Together, these layers make the overall approach more robust than any one safeguard on its own. Model behavior reduces the likelihood of harmful responses, real-time systems can intervene during generation, account-level review can identify broader patterns, and differentiated access preserves important defensive work without making the most sensitive capabilities broadly available by default.
Especially during the preview, users may encounter safeguards that block or refuse some requests. Other requests may take longer because generation is paused for additional review. Safeguards may occasionally intervene on legitimate work, particularly in dual-use areas where defensive and offensive activity can initially look similar.
That is part of what the preview is designed to test. We want to understand not only whether the safeguards constrain misuse, but whether legitimate users can still complete normal work reliably and efficiently. Feedback during the preview will help us reduce unnecessary blocks and delays, improve how the safeguards interpret context, and create a smoother experience before wider release.
We are also working with enterprise customers on longer-term approaches—including privacy-preserving detection, customer-operated safety controls, and access calibrated to the risk of a customer, user, or workload—to advance safety while supporting enterprise privacy requirements.
Zaštita koja djeluje samo na fiksni skup poznatih napada nije dovoljno robusna za napredni model.
Zato u sigurnost ulažemo više inteligencije i računalnih resursa nego ikad, koristeći vlastite modele kako bismo brže pronašli slabosti i poboljšali zaštitne mjere. Automatiziranom red-teamingu posvećenom pronalaženju univerzalnih jailbreakova namijenili smo više od 700.000 GPU sati ekvivalentnih A100: napada koji mogu funkcionirati u mnogim upitima ili kontekstima, a ne samo u jednom uskom okruženju. Usmjerenost na te teže, općenitije napade omogućila nam je da zaštitne mjere testiramo izvan fiksnog skupa poznatih neuspjeha. Omogućuje nam i da istražimo daleko više obrazaca napada nego što bi moglo pokriti samo ljudsko testiranje, ranije prepoznamo obrasce neuspjeha i skratimo put od pronalaska slabosti do njezina uklanjanja.
Uz automatizirani red-teaming, surađivali smo s vanjskim testerima na opsežnom red-teamingu koji provode ljudski stručnjaci, a koji će se nastaviti tijekom razdoblja pregleda. Ljudski red-teaming nadopunjuje automatizirani rad testiranjem zaštitnih mjera protiv kreativnih stručnjaka koji pokušavaju zloupotrijebiti model na načine koje naši sustavi možda ne bi predvidjeli.
Nijedna evaluacija ne može obuhvatiti svaku konfiguraciju proizvoda, višekoračni napad ili stvarni tijek rada. Zato održavamo postupak brzog odgovora kako bismo reproducirali, procijenili, odredili prioritet i otklonili novootkrivene jailbreakove, a zatim ih dodali u naše stalne evaluacije kako bismo ubuduće mogli testirati slične neuspjehe.
Tijekom pregleda modeli GPT‑5.6 u početku će biti dostupni putem API-ja i Codexa odabranoj skupini pouzdanih partnera i organizacija. Uskoro ih planiramo učiniti šire dostupnima korisnicima ChatGPT‑a, Codexa i API-ja.
U novom sustavu imenovanja uvedenom s GPT‑5.6 broj označava generaciju modela, dok Sol, Terra i Luna označavaju trajne razine sposobnosti koje mogu napredovati vlastitim ritmom. Ta obitelj zajedno korisnicima i razvojnim programerima pruža jasniji izbor prema inteligenciji, brzini i trošku.
GPT‑5.6 naplaćuje se po 1 mil. tokena za tri veličine modela: Sol je 5 USD za ulaz / 30 USD za izlaz; Terra 2,50 USD za ulaz / 15 USD za izlaz; a Luna 1 USD za ulaz / 6 USD za izlaz. GPT‑5.6 uvodi i predvidljivije predmemoriranje upita, uključujući podršku za izričite točke prekida predmemorije i minimalni vijek predmemorije od 30 minuta. Za GPT‑5.6 i kasnije modele, upisi u predmemoriju naplaćuju se po 1,25 puta višoj necachiranoj ulaznoj tarifi modela, dok čitanja iz predmemorije i dalje ostvaruju popust od 90 % za predmemorirani ulaz.
U srpnju pokrećemo i GPT‑5.6 Sol na Cerebrasu, s do 750 tokena u sekundi, donoseći korisnicima vrhunsku inteligenciju dosad nezabilježenom brzinom. Pristup će u početku biti ograničen na odabrane korisnike dok širimo kapacitete.
Veselimo se nastavku učenja tijekom ovog razdoblja pregleda i skorom približavanju modela GPT‑5.6 Sol, Terra i Luna većem broju ljudi.
1. Latenciju i trošak API-ja procjenjujemo promatrajući produkcijsko ponašanje svojih modela i simulirajući izvanmrežno. Te procjene uzimaju u obzir pojedinosti poziva alata, uzorkovane tokene i ulazne tokene. Rezultati u stvarnom svijetu mogu se znatno razlikovati i ovise o mnogim čimbenicima koji nisu obuhvaćeni našom simulacijom. Latenciju simuliramo pri velikim brzinama API-ja, a trošak prema redovitim cijenama API-ja.
2. Svi se modeli evaluiraju s pomoću API okvira ExploitBench s 5 početnih vrijednosti i kontinuitetom rasuđivanja.
3. ExploitGym smo pokrenuli na našem alfa API-ju, koji odgovore ispisuje brže od našeg javnog API-ja, a zatim smo rezultate preračunali kako bi odgovarali javnom API-ju. Pri preračunavanju latencija na brzine očekivane za naš javni API neke procijenjene latencije premašuju vremenska ograničenja od 2 h i 6 h, iako su u evaluacijskom pokretanju ispravno poštovana. Za veće brzine u vremenski osjetljivom radu nudimo prioritetnu obradu u API-ju i brzi način rada u Codexu.
4. Modeli bez prijavljenih izlaznih tokena, latencije ili troška prikazani su kao vodoravne točkaste linije.
