OpenAI has launched a limited preview of GPT-5.6, introducing a three-tier model family consisting of Sol, Terra, and Luna. The release organizes model performance around different workload requirements, with Sol serving as the flagship model, Terra delivering GPT-5.5-level performance at roughly half the cost, and Luna focusing on speed and affordability.
The announcement goes well past new model capabilities. GPT-5.6 introduces new reasoning modes while expanding cybersecurity research, scientific analysis, pricing flexibility, and the infrastructure that supports production AI environments.
Initial access is limited to a select group of trusted partners while OpenAI continues testing its newest safeguard systems and gathers operational feedback ahead of wider availability across ChatGPT, Codex, and the API in the coming weeks.
Why It Matters: Enterprise AI evaluation now extends well past benchmark performance. Long-term adoption depends just as much on governance, security, infrastructure, and operating economics as it does on model capability. GPT-5.6 brings those elements together in one release, offering a clearer view of how frontier AI is maturing for production use.
- Purpose-Built Model Tiers: Sol delivers OpenAI’s highest level of reasoning performance for demanding technical work. Terra provides performance comparable to GPT-5.5 while cutting costs in half, making it suitable for everyday business applications. Luna focuses on high-volume workloads where speed and affordability carry greater weight. The new naming convention establishes durable capability tiers that can evolve independently, giving organizations greater flexibility when selecting models for different operational needs.
- AI Moves Into Execution: GPT-5.6 introduces Max Reasoning along with Ultra mode, which coordinates specialized subagents to complete longer technical workflows with less manual coordination. OpenAI reports state-of-the-art performance across software engineering, genomics, quantitative biology, and cybersecurity benchmarks, with stronger results while using fewer tokens in several evaluations. These results show continued progress toward AI systems that complete extended technical work with less human intervention.
- Security by Design: GPT-5.6 launches with OpenAI’s largest safeguard stack, combining protections built into the model with real-time detection, account-level review, differentiated access, and ongoing evaluation. More than 700,000 A100-equivalent GPU hours supported automated red teaming before release, with external security experts expanding the testing effort. The layered approach seeks to preserve legitimate security research while reducing opportunities for prohibited misuse.
- Deployment With Guardrails: GPT-5.6 remains below OpenAI’s Cyber Critical threshold under its Preparedness Framework, even with stronger cybersecurity performance. The preview period gives OpenAI additional time to evaluate safeguards, reduce false positives, and improve system behavior before general availability. The company also previewed the models with the U.S. government before launch while work continues on a repeatable release framework.
- The Economics of AI: GPT-5.6 introduces updated API pricing along with prompt caching enhancements that improve cost predictability for production deployments. Cache reads continue to receive a 90% discount, while cache writes follow a new pricing structure beginning with GPT-5.6. OpenAI also plans to offer GPT-5.6 Sol on Cerebras at speeds of up to 750 tokens per second, giving organizations another option for high-performance inference.
Go Deeper -> Previewing GPT‑5.6 Sol: a next-generation model – OpenAI
Trusted insights for technology leaders
Our readers are CIOs, CTOs, and senior IT executives who rely on The National CIO Review for smart, curated takes on the trends shaping the enterprise, from GenAI to cybersecurity and beyond.
Subscribe to our 4x a week newsletter to keep up with the insights that matter.
