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(åbner i et nyt vindue). 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(åbner i et nyt vindue)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(åbner i et nyt vindue).
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.
Sikkerhedsforanstaltninger skal også blive ved med at virke, når angribere tilpasser deres taktik. En beskyttelse, der kun virker mod et fast sæt kendte angreb, er ikke robust nok til en frontier-model.
Derfor bruger vi mere intelligens og compute end nogensinde før på sikkerhed og anvender vores egne modeller til hurtigere at finde svagheder og forbedre sikkerhedsforanstaltninger. Vi afsatte over 700.000 A100-ækvivalente GPU-timer til automatiseret red-teaming med det formål at finde universelle jailbreaks: angreb, der kan fungere på tværs af mange prompts eller kontekster, ikke kun i én snæver situation. Ved at fokusere på disse sværere og mere generelle angreb kunne vi teste sikkerhedsforanstaltningerne ud over et fast sæt kendte fejl. Det lader os også undersøge langt flere angrebsmønstre, end menneskelig testning alene kunne dække, identificere fejlmønstre tidligere og forkorte vejen fra at finde en svaghed til at afhjælpe den.
Ud over automatiseret red-teaming samarbejdede vi med tredjepartstestere om omfattende red-teaming udført af menneskelige eksperter, og det fortsætter i forhåndsvisningsperioden. Menneskelig red-teaming supplerer det automatiserede arbejde ved at teste sikkerhedsforanstaltninger mod kreative eksperter, der forsøger at misbruge modellen på måder, vores systemer måske ikke forudser.
Ingen evaluering kan repræsentere alle produktkonfigurationer, flertrinsangreb eller arbejdsgange fra den virkelige verden. Derfor har vi en hurtig responsproces til at reproducere, vurdere, prioritere og afhjælpe nyopdagede jailbreaks og derefter føje dem til vores løbende evalueringer, så vi kan teste mod lignende fejl fremover.
I forhåndsvisningsperioden bliver GPT‑5.6-modellerne i første omgang tilgængelige via API'en og Codex for en udvalgt gruppe af betroede partnere og organisationer. Vi planlægger snart at gøre dem bredere tilgængelige for personer, der bruger ChatGPT, Codex og API'en.
I det nye navngivningssystem, der introduceres med GPT‑5.6, angiver tallet en models generation, mens Sol, Terra og Luna angiver varige kapabilitetsniveauer, der kan udvikle sig i deres eget tempo. Samlet giver familien brugere og udviklere klarere valg på tværs af intelligens, hastighed og omkostninger.
GPT‑5.6 prissættes pr. 1 mio. tokens på tværs af tre modelstørrelser: Sol koster 5 USD input/30 USD output; Terra koster 2,50 USD input/15 USD output; og Luna koster 1 USD input/6 USD output. GPT‑5.6 introducerer også mere forudsigelig prompt-caching, herunder understøttelse af eksplicitte cache-breakpoints og en cachelevetid på mindst 30 minutter. For GPT‑5.6 og senere modeller faktureres cacheskrivninger til 1,25x modellens inputpris uden cache, mens cachelæsninger fortsat får rabatten på 90 % for cachelagret input.
Vi lancerer også GPT‑5.6 Sol på Cerebras med op til 750 tokens pr. sekund i juli og giver kunder frontier-intelligens med hidtil uset hastighed. Adgangen vil i første omgang være begrænset til udvalgte kunder, mens vi udbygger kapaciteten.
Vi glæder os til fortsat at lære af denne forhåndsvisningsperiode og snart bringe GPT‑5.6 Sol, Terra og Luna ud til flere mennesker.
1. Vi estimerer latens og API-omkostninger ved at se på vores modellers produktionsadfærd og simulere offline. Disse estimater tager højde for detaljer om værktøjskald, samplede tokens og inputtokens. Resultater i den virkelige verden kan variere betydeligt og afhænger af mange faktorer, som ikke indgår i vores simulering. Vi simulerer latens ved hurtige API-hastigheder og omkostninger ved almindelige API-priser.
2. Alle modeller evalueres med ExploitBench API-harness med 5 seeds og ræsonneringskontinuitet.
3. Vi kørte ExploitGym på vores alpha-API, som outputter svar hurtigere end vores offentlige API, og skalerede derefter om, så det matchede vores offentlige API. Når latenser skaleres om til de hastigheder, der forventes for vores offentlige API, medfører det, at nogle estimerede latenser overstiger tidsgrænserne på 2 t og 6 t, selv om de blev overholdt korrekt i evalueringskørslen. For at få højere hastigheder til tidsfølsomt arbejde tilbyder vi prioriteret behandling i API'en og hurtig tilstand i Codex.
4. Modeller uden rapporterede outputtokens, latens eller omkostninger vises som vandrette stiplede linjer.
