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(mbukak ing jendhela anyar). 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(mbukak ing jendhela anyar)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(mbukak ing jendhela anyar).
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.
Pengaman uga kudu tetep efektif nalika panyerang ngganti taktiké. Perlindhungan sing mung bisa kanggo sakumpulan serangan sing wis dingerteni durung cukup tangguh kanggo model tercanggih.
Mula, saiki kita ngetrapake kecerdasan lan komputasi luwih gedhe tinimbang sadurunge kanggo safety, nggunakake model kita dhewe kanggo nemokake kelemahan lan luwih cepet ngapikake pengaman. Kita nyawisake luwih saka 700.000 jam GPU setara A100 kanggo red teaming otomatis sing ditujokake nemokake jailbreak universal: serangan sing bisa lumaku ing pirang-pirang prompt utawa konteks, dudu mung ing siji setelan sempit. Kanthi fokus marang serangan sing luwih angel lan luwih umum iki, kita bisa nguji pengaman ngluwihi daftar kegagalan sing wis dingerteni. Iki uga ngidini kita njelajah pola serangan luwih akeh tinimbang sing bisa dicakup uji coba manungsa wae, ngenali pola kegagalan luwih awal, lan nyepetake dalan saka nemokake kelemahan nganti ngatasi.
Saliyane red teaming otomatis, kita makarya karo penguji pihak katelu kanggo nindakake red teaming pakar manungsa kanthi jembar, sing bakal terus ditindakake sajrone periode pratinjau. Red teaming manungsa nglengkapi pakaryan otomatis kanthi nguji pengaman ngadhepi para pakar kreatif sing nyoba nyalahgunakake model kanthi cara sing bisa uga ora diantisipasi sistem kita.
Ora ana evaluasi sing bisa makili saben konfigurasi produk, serangan multi-langkah, utawa alur kerja donya nyata. Mula, kita njaga proses tanggap cepet kanggo ngasilake maneh, mbiji, nggawe prioritas, lan ndandani jailbreak anyar sing ditemokake, banjur nambahake menyang evaluasi terus-terusan supaya mbesuk bisa nguji kegagalan sing padha.
Sajrone pratinjau, model GPT‑5.6 wiwitane bakal kasedhiya liwat API lan Codex kanggo klompok mitra lan organisasi dipercaya sing kapilih. Kita ngrancang supaya model iki enggal kasedhiya luwih jembar kanggo wong sing nggunakake ChatGPT, Codex, lan API.
Ing sistem panamaan anyar sing dikenalake karo GPT‑5.6 iki, angka nuduhake generasi model, dene Sol, Terra, lan Luna nuduhake tingkat kemampuan awet sing bisa maju miturut irama dhewe. Bebarengan, kulawarga iki menehi pilihan sing luwih cetha kanggo wong lan pangembang babagan kecerdasan, kacepetan, lan biaya.
GPT‑5.6 diregani saben 1M token ing telung ukuran model: Sol $5 input / $30 output; Terra $2,50 input / $15 output; lan Luna $1 input / $6 output. GPT‑5.6 uga ngenalake caching prompt sing luwih bisa diprakirakake, kalebu dhukungan kanggo breakpoint cache eksplisit lan umur cache minimal 30 menit. Kanggo GPT‑5.6 lan model sabanjure, penulisan cache ditagih 1,25x tarif input tanpa cache saka model kasebut, dene maca cache tetep nampa diskon input cache 90%.
Kita uga ngluncurake GPT‑5.6 Sol ing Cerebras nganti 750 token per detik ing Juli, nggawa kecerdasan tercanggih marang pelanggan kanthi kacepetan sing durung ana sadurunge. Akses wiwitane bakal diwatesi kanggo pelanggan kapilih nalika kita nggedhekake kapasitas.
Kita bungah bisa terus sinau saka periode pratinjau iki, lan enggal nggawa GPT‑5.6 Sol, Terra, lan Luna marang luwih akeh wong.
1. Kita ngira latensi lan biaya API kanthi ndeleng perilaku produksi model kita, banjur nyimulasi kanthi offline. Perkiraan iki nyakup rincian panggilan alat, token sing disampel, lan token input. Asil donya nyata bisa beda banget, gumantung marang akeh faktor sing ora kacathet ing simulasi kita. Kita nyimulasi latensi ing kacepetan API cepet, lan biaya ing rega API reguler.
2. Kabeh model dievaluasi nganggo harness API ExploitBench kanthi 5 seed lan kesinambungan nalar.
3. Kita mbukak ExploitGym ing API alfa kita, sing ngasilake respons luwih cepet tinimbang API publik, banjur diskalakake maneh supaya cocog karo API publik. Nalika latensi diskalakake maneh menyang kacepetan sing dikarepake kanggo API publik, sawetara perkiraan latensi dadi ngluwihi wates wektu 2 jam lan 6 jam, sanajan ing run evaluasi sejatine wis manut kanthi bener. Kanggo entuk kacepetan luwih cepet kanggo pakaryan sing peka wektu, kita nyedhiyakake pangolahan prioritas ing API lan mode cepet ing Codex.
4. Model tanpa token output, latensi, utawa biaya sing dilaporake digambarake minangka garis titik-titik horizontal.
