OpenAI Unveils GPT-5.6 Sol, Terra, and Luna Amid U.S. Government Pre-Approval Controversy

모델 출시 | Sat Jun 27 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 7 sources

OpenAI unveiled three next-generation GPT-5.6 models but released them as a limited preview at the request of the Trump administration.

Analysis

[OpenAI] unveiled a three-model GPT-5.6 series in preview ^[1]^[2]^[4]

Lineup consists of flagship Sol
mid-tier Terra
and low-cost Luna
Terra is 2x cheaper than GPT-5.5
Luna delivers strong performance at the lowest cost

[GPT-5.6 Sol] strengthened agentic capabilities in coding, biology, and cybersecurity ^[1]^[4]

Achieved SOTA on Terminal-Bench 2.1
Outperformed GPT-5.5 on GeneBench v1 using fewer tokens
Matched Mythos Preview on ExploitBench using about 1/3 the tokens

[GPT-5.6 Sol] introduced max reasoning mode and ultra sub-agent mode ^[1]^[2]

max mode provides maximum reasoning effort for deeper reasoning
ultra mode leverages sub-agents to accelerate complex tasks
Sol pricing set at $5 per 1M input tokens and $30 per 1M output tokens

[Trump administration] restricted GPT-5.6 release and required user pre-approval ^[3]^[5]^[6]^[7]

Preview provided only to a small group of trusted partners
Each initial partner requires government approval
First U.S. company to receive a pre-release restriction order

[OpenAI] expressed opposition to prolonged government pre-approval processes ^[1]^[3]^[6]

Stated such government access processes should not become a long-term default
Accepted as a short-term measure while pursuing broader access
Working with the administration to develop a cyber executive order framework

[Anthropic] fully discontinued its model after a Fable 5 foreign-access block order ^[3]^[6]

Took the model offline following a White House export control directive
Even some of its own employees cannot use the latest model
Mythos 5 was also effectively banned by the Trump administration

Sources