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(odpre se v novem oknu). 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(odpre se v novem oknu)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(odpre se v novem oknu).
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.
Zaščitni ukrepi morajo ostati učinkoviti tudi, ko napadalci prilagajajo taktike. Zaščita, ki deluje le pri nespremenljivem naboru znanih napadov, ni dovolj robustna za napredni model.
Zato varnosti namenjamo več inteligence in računalniških virov kot kdaj koli prej ter z lastnimi modeli hitreje odkrivamo slabosti in izboljšujemo zaščitne ukrepe. Za samodejni red-teaming, namenjen iskanju univerzalnih jailbreakov, smo porabili več kot 700.000 ur GPU, enakovrednih A100: napadov, ki lahko delujejo v številnih pozivih ali kontekstih, ne le v enem ozkem okolju. Osredotočanje na te zahtevnejše, splošnejše napade nam je omogočilo, da zaščitne ukrepe preizkusimo širše od nespremenljivega nabora znanih odpovedi. Omogoča nam tudi raziskati veliko več vzorcev napadov, kot bi jih lahko zajelo zgolj človeško testiranje, prej prepoznati vzorce odpovedi in skrajšati pot od odkritja slabosti do njene odprave.
Poleg samodejnega red-teaminga smo s tretjimi preizkuševalci izvedli obsežen človeški strokovni red-teaming, ki se bo nadaljeval tudi v obdobju predogleda. Človeški red-teaming dopolnjuje samodejno delo, saj zaščitne ukrepe preizkuša proti ustvarjalnim strokovnjakom, ki poskušajo model zlorabiti na načine, ki jih naši sistemi morda ne bi predvideli.
Nobeno vrednotenje ne more zajeti vsake konfiguracije izdelka, večstopenjskega napada ali poteka dela v resničnem svetu. Zato vzdržujemo postopek hitrega odzivanja za reprodukcijo, oceno, prednostno razvrstitev in odpravo na novo odkritih jailbreakov, nato pa jih dodamo v svoja stalna vrednotenja, da lahko v prihodnje testiramo podobne odpovedi.
Med predogledom bodo modeli GPT‑5.6 sprva na voljo prek API-ja in Codexa izbrani skupini zaupanja vrednih partnerjev in organizacij. Kmalu jih nameravamo širše omogočiti uporabnikom ChatGPT, Codexa in API-ja.
V tem novem sistemu poimenovanja, uvedenem z GPT‑5.6, številka označuje generacijo modela, Sol, Terra in Luna pa trajne ravni zmogljivosti, ki lahko napredujejo v svojem ritmu. Družina skupaj uporabnikom in razvijalcem ponuja jasnejše izbire glede inteligence, hitrosti in stroškov.
GPT‑5.6 se obračunava na 1 milijon žetonov v treh velikostih modela: Sol je 5 USD za vhod / 30 USD za izhod; Terra je 2,50 USD za vhod / 15 USD za izhod; Luna pa 1 USD za vhod / 6 USD za izhod. GPT‑5.6 uvaja tudi bolj predvidljivo predpomnjenje pozivov, vključno s podporo za izrecne prekinitvene točke predpomnilnika in najmanjšo 30-minutno življenjsko dobo predpomnilnika. Pri GPT‑5.6 in poznejših modelih se zapisi v predpomnilnik obračunajo po 1,25-kratni nepredpomnjeni vhodni ceni modela, branja iz predpomnilnika pa še naprej prejmejo 90-odstotni popust za predpomnjeni vhod.
Julija uvajamo tudi GPT‑5.6 Sol na Cerebras s hitrostjo do 750 žetonov na sekundo, kar strankam prinaša napredno inteligenco z doslej neprimerljivo hitrostjo. Dostop bo sprva omejen na izbrane stranke, medtem ko širimo zmogljivosti.
Veselimo se nadaljnjega učenja v tem obdobju predogleda in tega, da bomo GPT‑5.6 Sol, Terra in Luna kmalu ponudili več ljudem.
1. Latenco in strošek API-ja ocenjujemo na podlagi produkcijskega vedenja svojih modelov ter s simulacijo brez spletne povezave. Te ocene upoštevajo podrobnosti klicev orodij, vzorčene žetone in vhodne žetone. Rezultati v resničnem svetu se lahko bistveno razlikujejo in so odvisni od številnih dejavnikov, ki jih naša simulacija ne zajame. Latenco simuliramo pri hitrih hitrostih API-ja, strošek pa pri rednih cenah API-ja.
2. Vsi modeli so ovrednoteni z ogrodjem API ExploitBench s 5 semeni in kontinuiteto sklepanja.
3. ExploitGym smo izvajali na svojem alfa API-ju, ki odgovore vrača hitreje kot naš javni API, nato pa rezultate preračunali, da ustrezajo javnemu API-ju. Pri preračunavanju latenc na hitrosti, pričakovane za naš javni API, nekatere ocenjene latence presežejo časovni omejitvi 2 h in 6 h, čeprav so bile v vrednotenju pravilno upoštevane. Za večje hitrosti pri časovno občutljivem delu v API-ju ponujamo prednostno obdelavo, v Codex pa hitri način.
4. Modeli brez poročanih izhodnih žetonov, latence ali stroška so prikazani kot vodoravne pikčaste črte.
