OpenAI on Thursday launched its GPT-5.6 family of artificial intelligence models, Sol, Terra and Luna, marking the general availability of its latest frontier AI models following a limited preview and a delayed broader rollout after a security review with the U.S. government. Alongside the launch, the company compared its flagship GPT-5.6 Sol with Anthropic’s Claude Fable 5, highlighting gains in performance, speed and cost efficiency.
The GPT-5.6 family includes three models tailored for different use cases. GPT-5.6 Sol, the flagship model, is designed for advanced reasoning, coding, scientific research and cybersecurity tasks. GPT-5.6 Terra balances performance, speed and cost for enterprise and developer workloads, while GPT-5.6 Luna is optimized for high-speed, low-cost AI applications.
OpenAI also introduced Ultra, a new multi-agent reasoning mode that enables multiple AI agents to work together on complex tasks.
The broader rollout of GPT-5.6 was delayed while the model went through a security review involving U.S. government officials amid concerns over the potential national security implications of increasingly capable AI systems. During the review, OpenAI provided preview access to a limited group of trusted U.S.-based partners.
According to OpenAI, GPT-5.6 was designed to deliver more useful work from every token while reducing inference costs. The company said GPT-5.6 Sol scored 53.6 on Agents’ Last Exam, a benchmark measuring long-running professional workflows across 55 fields, points higher higher than Anthropic’s Claude Fable 5 by 13.1 points. OpenAI said Sol also exceeded Claude Fable 5 by 11.4 points in its medium reasoning mode at roughly one-quarter of the estimated cost.
According to OpenAI, GPT-5.6 Sol is capable of “outperforming previous and competing frontier models with fewer tokens and at lower estimated cost.”
On the Artificial Analysis Intelligence Index, GPT‑5.6 Sol with max reasoning comes within one point of Fable 5 while completing tasks in 61% less time at roughly half the estimated cost, the company said.
The company added that the advantage extends across the family: Terra performs just above Fable 5, while Luna outperforms Opus 4.8. According to OpenAI, the GPT-5.6 family also posted leading results on Terminal Bench 2.1 and DeepSWE, which evaluate complex software engineering workflows.
GPT-5.6 is rolling out across ChatGPT, Codex and the OpenAI API, with global availability expanding over the next 24 hours.
The launch comes as competition in the AI industry intensifies, with companies including OpenAI, Anthropic, xAI, Google and Meta racing to develop increasingly capable frontier AI models amid growing government scrutiny over their safety and national security implications.
Also Read: Global Leaders Intensify Debate Over Access to Advanced AI Models Amid U.S. Restrictions
