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(terbuka di jendela baru). 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(terbuka di jendela baru)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(terbuka di jendela baru).
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 juga harus tetap efektif ketika penyerang menyesuaikan taktik mereka. Perlindungan yang hanya berfungsi pada sekumpulan serangan yang sudah dikenal tidak cukup tangguh untuk model frontier.
Karena itu, kami menerapkan lebih banyak kecerdasan dan komputasi daripada sebelumnya untuk keselamatan, menggunakan model kami sendiri guna menemukan kelemahan dan memperbaiki pengaman lebih cepat. Kami mendedikasikan lebih dari 700.000 jam GPU setara A100 untuk red-teaming otomatis yang bertujuan menemukan jailbreak universal: serangan yang dapat bekerja di banyak prompt atau konteks, bukan hanya satu skenario sempit. Dengan berfokus pada serangan yang lebih sulit dan lebih umum ini, kami dapat menguji pengaman melampaui sekumpulan kegagalan yang sudah dikenal. Ini juga memungkinkan kami menjelajahi jauh lebih banyak pola serangan daripada yang dapat dicakup pengujian manusia saja, mengidentifikasi pola kegagalan lebih awal, dan memperpendek jarak dari menemukan kelemahan hingga mengatasinya.
Selain red-teaming otomatis, kami bekerja dengan penguji pihak ketiga untuk melakukan red-teaming pakar manusia secara ekstensif, yang akan berlanjut selama periode pratinjau. Red-teaming oleh manusia melengkapi pekerjaan otomatis dengan menguji pengaman terhadap pakar kreatif yang mencoba menyalahgunakan model dengan cara yang mungkin tidak diantisipasi sistem kami.
Tidak ada evaluasi yang dapat mewakili setiap konfigurasi produk, serangan multi-langkah, atau alur kerja dunia nyata. Karena itu, kami mempertahankan proses respons cepat untuk mereproduksi, menilai, memprioritaskan, dan memperbaiki jailbreak yang baru ditemukan, lalu menambahkannya ke evaluasi berkelanjutan kami agar dapat menguji kegagalan serupa di masa mendatang.
Selama pratinjau, model GPT‑5.6 awalnya akan tersedia melalui API dan Codex bagi sekelompok mitra dan organisasi tepercaya yang dipilih. Kami berencana segera membuatnya tersedia lebih luas bagi orang-orang yang menggunakan ChatGPT, Codex, dan API.
Dalam sistem penamaan baru yang diperkenalkan bersama GPT‑5.6 ini, angka menunjukkan generasi model, sementara Sol, Terra, dan Luna menunjukkan tingkat kemampuan yang tahan lama dan dapat berkembang dengan ritmenya sendiri. Secara keseluruhan, keluarga ini memberi pengguna dan developer pilihan yang lebih jelas dalam hal kecerdasan, kecepatan, dan biaya.
GPT‑5.6 diberi harga per 1 juta token di tiga ukuran model: Sol seharga $5 input / $30 output; Terra seharga $2,50 input / $15 output; dan Luna seharga $1 input / $6 output. GPT‑5.6 juga memperkenalkan caching prompt yang lebih dapat diprediksi, termasuk dukungan untuk breakpoint cache eksplisit dan masa cache minimum 30 menit. Untuk GPT‑5.6 dan model berikutnya, penulisan cache ditagih 1,25x tarif input tanpa cache model tersebut, sementara pembacaan cache tetap menerima diskon input-cache 90%.
Kami juga meluncurkan GPT‑5.6 Sol di Cerebras dengan kecepatan hingga 750 token per detik pada Juli, menghadirkan kecerdasan frontier kepada pelanggan dengan kecepatan yang belum pernah ada sebelumnya. Akses awalnya akan dibatasi untuk pelanggan tertentu sementara kami memperluas kapasitas.
Kami antusias untuk terus belajar dari periode pratinjau ini, dan segera menghadirkan GPT‑5.6 Sol, Terra, dan Luna kepada lebih banyak orang.
1. Kami memperkirakan latensi dan biaya API dengan melihat perilaku produksi model kami, lalu menyimulasikannya secara offline. Perkiraan ini memperhitungkan detail panggilan alat, token yang disampel, dan token input. Hasil di dunia nyata dapat sangat berbeda, dan bergantung pada banyak faktor yang tidak tercakup dalam simulasi kami. Kami menyimulasikan latensi pada kecepatan API cepat, dan biaya pada harga API reguler.
2. Semua model dievaluasi menggunakan harness API ExploitBench dengan 5 seed dan kontinuitas penalaran.
3. Kami menjalankan ExploitGym pada API alpha kami, yang menghasilkan respons lebih cepat daripada API publik kami, lalu menskalakannya ulang agar sesuai dengan API publik kami. Saat menskalakan ulang latensi ke kecepatan yang diharapkan untuk API publik kami, sebagian estimasi latensi melebihi batas waktu 2 jam dan 6 jam, meskipun batas tersebut dipatuhi dengan benar dalam run evaluasi. Untuk memperoleh kecepatan lebih tinggi bagi pekerjaan sensitif waktu, kami menawarkan pemrosesan prioritas di API dan mode cepat di Codex.
4. Model tanpa token output, latensi, atau biaya yang dilaporkan ditampilkan sebagai garis putus-putus horizontal.
