In a consequential leap for the artificial intelligence landscape, OpenAI has officially unveiled its GPT-5.6 family of models to the public.

The GPT-5.6 Trinity: Sol, Terra, and Luna

The rollout introduces three discrete tiers designed to address varying computational demands. GPT-5.6 Sol stands as the premier flagship, engineered for the most arduous agentic tasks, boasting an unprecedented 91.9% score on Terminal-Bench 2.1 in ultra mode. GPT-5.6 Terra offers a balanced paradigm for enterprise workflows, delivering near-GPT-5.5 performance at a fraction of the cost. Meanwhile, GPT-5.6 Luna serves as the most economical iteration, optimized for high-volume, latency-sensitive applications.

"GPT-5.6 Sol sets a new standard for both intelligence and efficiency, achieving state-of-the-art results across coding, knowledge work, cybersecurity, and science while outperforming previous and competing frontier models with fewer tokens and at lower estimated cost." — OpenAI Official Announcement

Economic and Architectural Shifts

The pricing strategy fundamentally alters the enterprise AI economics. Terra is priced at $2.50 per million input tokens and $15 output, making it an extremely attractive proposition for businesses previously routing to open-weight models. Sol commands $5 input and $30 output, justified by its unmatched capability in long-horizon software engineering. Furthermore, the introduction of Programmatic Tool Calling allows the model to write and run lightweight programs in-memory, significantly reducing token consumption and model round trips.

Unprecedented Safety Protocols

OpenAI has implemented its most thorough safety stack to date. The GPT-5.6 models underwent approximately 700,000 A100e GPU hours of black-box automated red teaming. The safeguards are layered, combining protections trained into the model with real-time checks and a reasoning monitor that evaluates the conversation context to determine potential harm, ensuring defensive work is preserved while blocking serious misuse.

Global Availability

The models are currently being deployed globally across ChatGPT, Codex, and the OpenAI API. The rollout is progressive, with paid subscribers seeing the new tiers in their model selector dropdowns over the next 24 to 48 hours. API developers must explicitly update their model strings to gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna, as the gpt-5.5-latest endpoint will not auto-migrate.