OpenAI Unveils New AI Model 'GPT-5.6', Highlights 'Sol' Performance

OpenAI has launched its new AI model 'GPT-5.6'. OpenAI stated that the 'Sol' model possesses its best-ever performance and autonomous agent capabilities in coding, biology, and cybersecurity. The model newly introduces a 'maximum reasoning effort' option for rapidly executing complex tasks and an 'Ultra Mode'. It achieved a record of 91.9% on the 'Terminal-Bench 2.1' performance metric, surpassing competitor Claude Mitos5.

The 'Sol' model is equipped with a strengthened safety framework that includes enhanced safeguards against high-risk activities or concerns of hacking exploitation.

The mid-tier model, 'Terra', offers a reduced price, while the lower-tier model, 'Luna', provides decent performance at the lowest cost.

These models will initially be provided to 'trusted partners', with general distribution expected in a few weeks. OpenAI pointed out that this form of prioritized access, following government requests, should not become a long-term standard. OpenAI did not disclose detailed performance metrics for the new model at its announcement.

Meanwhile, it has been reported that OpenAI is considering postponing its initial public offering (IPO) plans, which were originally anticipated for this year, to next year.