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(ku furmaa daaqad cusub). 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(ku furmaa daaqad cusub)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(ku furmaa daaqad cusub).
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.
Ilaalintu sidoo kale waa inay sii ahaataa mid wax-ku-ool ah marka weeraryahannadu la qabsadaan xeeladahooda. Ilaalin shaqaysa oo keliya marka la eego urur go’an oo weerarro la yaqaan ah ma aha mid ku filan nooc ugu casriyeysan.
Taasi waa sababta aan u adeegsanayno caqli iyo awood xisaabeed ka badan sidii hore dhinaca badbaadada, annagoo adeegsanayna nooc-yadeenna si aan u helno daldaloolo oo aan u dardargelinno hagaajinta ilaalinta. Waxaan u qoondeynay in ka badan 700,000 saacadood oo GPU ah oo u dhigma A100 si loogu sameeyo qiimeynta awoodda nidaamka ee otomaatiga ah, taas oo lagu beegsanayo helidda xadgudubyada hareer marka xeerarka badbaadada ee nidaam: weerarro ka shaqayn kara weydiinno ama duruufo badan, ee aan ku koobnayn hal xaalad oo cidhiidhi ah. Diiradda saarista weerarradan adag ee guud waxay noo oggolaatay inaan tijaabino ilaalinta ka baxsan urur go’an oo guuldarrooyin la yaqaan ah. Waxay sidoo kale noo oggolaanaysaa inaan sahaminno qaabab weerar oo aad uga badan inta tijaabada bini’aadamka keliya dabooli karto, inaan si hore u aqoonsanno qaababka guuldarrada, iyo inaan gaabinno jidka u dhexeeya helidda daldalool iyo wax ka qabashadiisa.
Marka laga soo tago qiimeynta awoodda nidaamka ee otomaatiga ah, waxaan la shaqaynay tijaabiyeyaal dhinac saddexaad ah si loo sameeyo qiimeyn ballaaran oo khubaro bini’aadam ah, taas oo sii socon doonta inta lagu jiro muddada horudhaca. Qiimeynta bini’aadamku waxay dhammaystirtaa shaqada otomaatiga ah iyadoo tijaabinaysa ilaalinta marka ay khubaro hal-abuur leh isku dayaan inay si khaldan u adeegsadaan nooca siyaabo ay nidaamyadeennu uusan filan karin.
Qiimeyn kastaa ma matali karto qaabeyn kasta oo badeecad, weerar tallaabooyin badan leh, ama hab-socod shaqo oo dunida dhabta ah. Sidaa darteed waxaan haynaa hannaan jawaab-degdeg ah oo dib loogu soo saaro, lagu qiimeeyo, lagu kala mudnaan siiyo, laguna saxo weerara jabinta xeerarka badbaadada ee nidaamyada ee cusub ee la ogaado; dabadeedna waxaan ku darnaa qiimeynteena socda si aan mustaqbalka uga tijaabino guuldarrooyin la mid ah.
Inta lagu jiro horudhaca, nooc-yada GPT‑5.6 marka hore waxa laga heli doonaa API-ga iyo Codex, waxaana loo heli doonaa koox xaddidan oo lammaane iyo hay’ado la aamini karo. Waxaan qorshaynaynaa inaan si ballaaran ugu fidinno dadka isticmaala ChatGPT, Codex, iyo API-ga dhawaan.
Nidaamkan magac-bixinta cusub ee lagu soo bandhigay GPT‑5.6, lambarku wuxuu tilmaamayaa jiilka nooc, halka Sol, Terra, iyo Luna ay tilmaamayaan heerar awood oo waara oo horumari kara jadwalkooda u gaarka ah. Wadar ahaan, qoyska noocyadu wuxuu dadka iyo horumariyeyaasha siinayaa doorashooyin ka sii cad oo ku saabsan caqli, xawaare, iyo kharash.
GPT‑5.6 waxaa lagu qiimeeyaa 1M qeybo-qoraal halkii, iyadoo leh saddex cabbir oo nooc ah: Sol waa $5 gelin / $30 soo-saar; Terra waa $2.50 gelin / $15 soo-saar; Luna-na waa $1 gelin / $6 soo-saar. GPT‑5.6 sidoo kale wuxuu soo bandhigayaa kaydin weydiin oo la sii saadaalin karo, oo ay ku jirto taageero loogu talagalay meelaha si cad loo jaro kaydka iyo ugu yaraan 30-daqiiqo cimri kayd. GPT‑5.6 iyo nooc-yada ka dambeeya, qorista kaydka waxaa lagu dallacaa 1.25x heerka gelinta aan kaydsanayn ee nooca, halka akhriska kaydka uu sii wato helidda dhimista 90% ee gelinta kaydsan.
Waxaan sidoo kale bilaabaynaa GPT‑5.6 Sol oo ku shaqaynaya Cerebras ilaa 750 qeybo-qoraal ilbiriqsigii bisha Luulyo, annagoo u keenayna caqli ugu casriyeysan macaamiisha xawaare aan hore loo arag. Helitaanku marka hore wuxuu ku koobnaan doonaa macaamiil la xushay inta aan ballaarinayno awoodda.
Waxaan ku faraxsanahay inaan sii wadno barashada muddadan horudhaca ah, isla markaana aan dhowaan u gaarsiino GPT‑5.6 Sol, Terra iyo Luna dad badan.
1. Waxaan qiyaasnaa dib-u-dhaca iyo kharashka API-ga annagoo eegayna hab-dhaqanka wax-soo-saarka ee nooc-yadeenna, kuna samaynayna jilid ofleen ah. Qiyaasahani waxay tixgeliyaan faahfaahinta wicitaanka qalabka, qeybo-qoraallada la muunadeeyay, iyo qeybo-qoraallada gelinta. Natiijooyinka dunida dhabta ah aad bay u kala duwanaan karaan, waxayna ku xiran yihiin arrimo badan oo aan ku jirin jiliddeena. Waxaan jilnaa dib-u-dhaca xawaareyaal API oo degdeg ah, kharashkana qiimeynta API ee caadiga ah.
2. Dhammaan nooc-yada waxaa lagu qiimeeyaa iyadoo la adeegsanayo ExploitBench API harness oo leh 5 setinno aan kala sooc lahayn iyo joogteynta caqliyeynta.
3. Waxaan ku socodsiinnay ExploitGym API-gayaga noociisa hore, kaas oo soo saara jawaabo ka dheereeya API-gayaga dadweynaha, dabadeedna waan dib-u-miisaannay si uu ula jaanqaado API-gayaga dadweynaha. Marka dib-u-dhacyada dib loogu miisaamo xawaareyaasha la filayo ee API-gayaga dadweynaha, tani waxay keeni kartaa in qaar ka mid ah dib-u-dhacyada la qiyaasay ay ka bataan xadka 2h iyo 6h saacadood, inkastoo si sax ah loo raacay intii lagu jiray orodka qiimeynta. Si loo helo xawaare degdeg ah shaqooyinka waqtiga xasaasiga ah, waxaan ku bixinnaa habayn mudnaan leh gudaha API-ga iyo hab degdeg ah gudaha Codex.
4. Nooc-yada aan laga soo sheegin qeybo-qoraallada soo-saar, dib-u-dhac ama kharash waxaa lagu muujinayaa xariiqyo dhibco ah oo jiif ah.
