OpenAI Twitter Translation.
https://x.com/OpenAI/status/2070555272230384038
Introducing GPT-5.6 Sol, a limited preview of our next-generation frontier model, GPT-5.6 Terra, a balanced model for efficient everyday tasks, and GPT-5.6 Luna, a fast and cost-effective model for large workloads.
We aim for broad accessibility and plan to make GPT-5.6 Sol, Terra, and Luna generally available in the coming weeks. For now, we are starting with a limited preview for a select few trusted partners through Codex and our API, in accordance with a request from the US government.
Sol is our new flagship model and represents a significant leap forward from GPT-5.5.
Terra offers performance comparable to GPT-5.5 at twice the cost savings.
Luna is our most cost-effective model, delivering powerful performance at the lowest cost.
Overall, the GPT-5.6 product line gives users and developers more choices to balance intelligence, speed, and cost.
GPT-5.6 Sol achieved a new state-of-the-art performance on Terminal-Bench 2.1, which evaluates complex command-line workflows requiring planning, repetitive tasks, and tool orchestration.
GPT-5.6 Sol boasts the most powerful performance we've seen from our models in the cybersecurity domain.
This model pushes the boundaries of performance and efficiency when performing long-term security tasks, including vulnerability research and exploitation.
GPT-5.6 Sol launches with the most robust safety stack we've ever deployed.
We have strengthened real-time safeguards against high-risk cyber activity and repeated misuse, followed by weeks of system hardening through red team exercises involving human participation and over 700,000 hours of A100 GPU time in automated testing.
