Previewing GPT‑5.6 Sol: a next-generation model
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(se abre en una ventana nueva). 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(se abre en una ventana nueva)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(se abre en una ventana nueva).
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.
Las medidas de seguridad también deben seguir siendo eficaces cuando los atacantes adaptan sus tácticas. Una protección que solo funciona frente a un conjunto fijo de ataques conocidos no es lo bastante robusta para un modelo de frontera.
Por eso destinamos más inteligencia y capacidad de cálculo que nunca a la seguridad y utilizamos nuestros propios modelos para detectar vulnerabilidades y reforzar las medidas de seguridad con mayor rapidez. Hemos dedicado el equivalente a más de 700 000 horas de GPU A100 a pruebas automatizadas de red teaming diseñadas para encontrar jailbreaks universales, es decir, ataques capaces de funcionar en una gran variedad de prompts y contextos, y no solo en un caso concreto.Al centrarnos en estos ataques más complejos y generalizables, hemos podido poner a prueba nuestras medidas de seguridad más allá de un conjunto limitado de fallos conocidos. Este enfoque también nos permite explorar muchos más patrones de ataque de los que podrían abarcar las pruebas realizadas únicamente por personas, detectar antes los patrones de fallo y acortar el tiempo que transcurre entre la detección de una vulnerabilidad y su corrección.
Además del red teaming automatizado, trabajamos con evaluadores externos para llevar a cabo un red teaming amplio con expertos humanos, que continuará durante el periodo de vista previa. El red teaming humano complementa el trabajo automatizado al poner a prueba las medidas de seguridad frente a expertos creativos que intentan hacer un uso indebido del modelo de formas que nuestros sistemas quizá no anticipen.
Ninguna evaluación puede representar todas las configuraciones de producto, ataques de varios pasos o flujos de trabajo reales. Por ello mantenemos un proceso de respuesta rápida para reproducir, evaluar, priorizar y remediar los jailbreaks recién descubiertos, y después añadirlos a nuestras evaluaciones continuas para poder probar fallos similares en el futuro.
Durante la vista previa, los modelos GPT‑5.6 estarán disponibles inicialmente a través de la API y Codex para un grupo selecto de socios y organizaciones de confianza. Tenemos previsto ponerlos pronto a disposición de más personas que usan ChatGPT, Codex y la API.
En este nuevo sistema de nombres introducido con GPT‑5.6, el número identifica la generación de un modelo, mientras que Sol, Terra y Luna identifican niveles de capacidad duraderos que pueden avanzar a su propio ritmo. En conjunto, la familia ofrece a personas y desarrolladores opciones más claras en inteligencia, velocidad y coste.
GPT‑5.6 tiene precios por 1 M de tokens en tres tamaños de modelo: Sol cuesta 5 $ de entrada / 30 $ de salida; Terra, 2,50 $ de entrada / 15 $ de salida; y Luna, 1 $ de entrada / 6 $ de salida. GPT‑5.6 también introduce un almacenamiento en caché de prompts más predecible, con compatibilidad para puntos de interrupción de caché explícitos y una vida mínima de caché de 30 minutos. Para GPT‑5.6 y modelos posteriores, las escrituras en caché se facturan a 1,25 veces la tarifa de entrada sin caché del modelo, mientras que las lecturas en caché siguen recibiendo el descuento del 90 % para entrada en caché.
También lanzaremos GPT‑5.6 Sol en Cerebras con hasta 750 tokens por segundo en julio, llevando inteligencia de frontera a los clientes a una velocidad sin precedentes. El acceso se limitará inicialmente a clientes seleccionados mientras ampliamos la capacidad.
Nos entusiasma seguir aprendiendo de este periodo de vista previa y llevar pronto GPT‑5.6 Sol, Terra y Luna a más personas.
1. Estimamos la latencia y el coste de API observando el comportamiento en producción de nuestros modelos y simulando sin conexión. Estas estimaciones tienen en cuenta los detalles de las llamadas a herramientas, los tokens muestreados y los tokens de entrada. Los resultados reales pueden variar sustancialmente y dependen de muchos factores que no se recogen en nuestra simulación. Simulamos la latencia a velocidades rápidas de API y el coste con los precios habituales de la API.
2. Todos los modelos se evalúan usando el arnés de la API de ExploitBench con 5 semillas y continuidad de razonamiento.
3. Ejecutamos ExploitGym en nuestra API alfa, que genera respuestas más rápido que nuestra API pública, y después reescalamos para igualar nuestra API pública. Al reescalar las latencias a las velocidades esperadas para nuestra API pública, algunas latencias estimadas superan los límites de tiempo de 2 h y 6 h, aunque se respetaron correctamente en la ejecución de la evaluación. Para obtener velocidades más rápidas en trabajos sensibles al tiempo, ofrecemos procesamiento prioritario en la API y modo rápido en Codex.
4. Los modelos sin tokens de salida, latencia o coste comunicados se muestran como líneas horizontales de puntos.
