OpenAI has released a limited preview of its GPT-5.6 model family, introducing three distinct models — Sol, Terra, and Luna — designed to offer users flexibility across intelligence, speed, and price. The announcement marks a strategic shift toward tiered AI offerings, with pricing ranging from $1 to $30 per million tokens depending on the model and task.
The Three Models and Their Pricing
GPT-5.6 Sol is positioned as the flagship model, delivering the highest performance in coding, biological research, and cybersecurity tasks. According to OpenAI, Sol sets new records on Terminal-Benchmark 2.1 for command-line workflows, GeneBench v1 for genomics, and the ExploitGym cybersecurity benchmark. Pricing for Sol is set at $5 per million input tokens and $30 per million output tokens.
GPT-5.6 Terra is a hybrid model that matches the performance of GPT-5.5 on everyday tasks while offering a 50% cost reduction. It is priced at $2.50 for input and $15 for output per million tokens. GPT-5.6 Luna is the fastest and cheapest option, designed for high-volume, lower-stakes applications, with input costs of $1 and output costs of $6 per million tokens. All three models show improved task completion as reasoning effort increases, particularly in cybersecurity benchmarks.
Availability and Safety Measures
OpenAI is initially offering the GPT-5.6 family through a limited preview of Codex and its API, restricted to selected partners approved by the U.S. government. The company plans to broaden access to Codex, the API, and ChatGPT in the coming weeks, though it emphasized that government-controlled access should not become the long-term norm.
The models launch with what OpenAI calls its most comprehensive safety stack to date, including enhanced protections against high-risk cyber activities and repeated misuse. Real-time classifiers will temporarily pause generation for review if they detect potentially harmful behavior. OpenAI confirmed that Sol did not exceed the critical threshold in its Preparedness Framework for cyber risk assessment.