OpenAI has officially introduced a new family of models, GPT-5.6, which includes three variants: the flagship Sol, the versatile Terra, and the lightweight Luna. For now, access via API and Codex is only open to a limited circle of trusted partners — a preview organized at the request of U.S. authorities. The company emphasizes that this format will not become the norm, and the models will become publicly available in the coming weeks.

Hierarchy of Capabilities and Pricing Policy

In the new lineup, Sol holds the position of the most powerful model in OpenAI's history. It is equipped with the max reasoning mode, which allocates more time for deep processing of complex tasks, and the ultra mode, which uses sub-agents to accelerate multi-step operations. Terra is positioned as a "workhorse" for everyday tasks — according to the developers, it is comparable in performance to GPT-5.5 but costs half as much. Luna is the fastest and most affordable option, ideal for simple queries.

Pricing reflects this hierarchy: for 1 million input tokens, Sol costs $5, Terra $2.50, and Luna $1; output tokens cost $30, $15, and $6, respectively.

Performance Records and Safety

According to OpenAI's internal tests, Sol set a new record in the Terminal-Bench 2.1 benchmark for command-line tasks. On GeneBench v1, it outperformed GPT-5.5 with fewer tokens, and on ExploitBench, it showed results comparable to Mythos Preview while using only a third of the output tokens. In ExploitGym tests, all three models demonstrated improved results in cyber tasks as reasoning depth increased.

Safety is a key focus of this release. OpenAI states that GPT-5.6 Sol does not exceed the Cyber Critical threshold within the Preparedness Framework. In tests with Chromium and Firefox, the model found bugs and exploit primitives but could not autonomously create a fully functional working exploit. Over 700,000 GPU hours (in A100 equivalent) were used for automated red teaming to verify protections. During the preview phase, multi-layered measures are in place: model-level restrictions, real-time checks, account-level signals, and monitoring.

Future Plans

In July, OpenAI plans to launch GPT-5.6 Sol on the Cerebras platform with speeds of up to 750 tokens per second, but access at this stage will also be limited. The company continues to expand its capacity. Notably, in early June, OpenAI confidentially filed for an IPO with the U.S. Securities and Exchange Commission — a signal of the company's maturity and readiness for the public market.

Expert Commentary: The choice of three models — Sol, Terra, Luna — is not just a marketing move but a strategic market segmentation. Sol targets research and cybersecurity tasks where price is irrelevant, Terra focuses on corporate clients with high performance demands, and Luna aims at the mass market and startups. However, limited access at the request of U.S. authorities raises questions about AI regulation: if the government is already dictating preview conditions, what will happen when the models become publicly available? The market should prepare for stricter oversight.