OpenAI's GPT-5.6 trio: Sol, Terra, Luna launch amid US security review
Three models, 50% cheaper than Anthropic, and a government-mandated preview period.
OpenAI unveiled GPT-5.6 on Friday, a model suite comprising Sol (flagship), Terra (high-volume), and Luna (fast/affordable). Priced at $5 input/$30 output per million tokens, Sol undercuts Anthropic's Claude Fable 5 by nearly half. The models are specifically tuned for coding, cybersecurity, and biology, and feature enhanced long-horizon agentic capabilities. OpenAI introduced two additional reasoning modes for Sol: 'max' for deeper analysis and 'ultra' for leveraging sub-agents, hinting at OpenClaw creator Peter Steinberger's influence.
The launch comes less than 24 hours after news that the Trump administration requested a staggered release, leading to a limited preview where customers are approved case-by-case. OpenAI dedicated most of its announcement to safety, noting 700,000 A100e GPU hours in automated red-teaming and third-party testing. The company emphasized that Sol 'is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks' and doesn't cross its revised preparedness framework's critical threshold. OpenAI hopes this government intervention won't set a precedent, stating they 'don't believe this kind of government access process should become the long-term default.' General availability is expected in weeks.
- GPT-5.6 includes three models: Sol (flagship, $5/$30 per million tokens), Terra (medium), and Luna (budget).
- Sol is nearly half the cost of Anthropic's Claude Fable 5 ($10/$50) and excels at coding, cybersecurity, and biology.
- Launch follows Trump admin request for case-by-case customer approval; OpenAI spent 700K A100e GPU hours on safety testing.
Why It Matters
GPT-5.6 sets a new price-performance bar while navigating unprecedented US government oversight of AI releases.