OpenAI has made GPT-5.6 Sol available to the public and has simultaneously launched two more affordable models, Terra and Luna. Previously, this version was restricted by the U.S. Department of Commerce and only offered for a two-week preview to approximately 20 trusted partners.
First-time use of named hierarchy
This is the first time OpenAI has deviated from purely numeric naming, instead using Sol, Terra, and Luna to distinguish capability tiers. According to the company, Sol is the flagship version; Terra is designed for everyday use, offering performance close to GPT-5.5 at roughly half the price; and Luna is a more affordable option.
OpenAI has also adjusted its invocation method. Sol offers a "Maximum Reasoning Intensity" option to extend thinking time, and introduces an "Ultra Mode" that distributes tasks among multiple sub-agents for collaborative processing.
- Sol is priced at $5 per million input tokens and $30 per million output tokens.
- Luna is priced at $1 per million input tokens and $6 per million output tokens.
- In comparison, Claude Fable 5 is $10/$50, while Grok 4.5 is $15/$75.
Benchmarking targets high-end models.
OpenAI has focused on performance comparisons. The company states that Sol achieved a score of 91.9% on Terminal-Bench 2.1 in ultra mode, surpassing Claude Mythos 5, Claude Fable 5, Claude Opus 4.8, and Gemini 3.1 Pro Preview.
In cybersecurity tests such as ExploitBench, OpenAI stated that Sol’s performance matched Anthropic’s restricted preview, Mythos Preview, while using approximately one-third the number of tokens. The company also noted that Sol has not yet reached the “Cyber Critical” level within its internal risk framework.
Competition among large models continues to accelerate
This release coincides with a wave of product updates from leading U.S. AI companies. The day before OpenAI launched GPT-5.6 Sol, xAI released Grok 4.5; hours later, Meta rolled out its paid model Muse Spark 1.1. Reports note that, in this round of updates, Google has become the major lab yet to refresh its flagship model, still relying on Gemini 3 since its November 2025 release.
Feedback from early testers also indicates that the market is viewing Sol as a new leading model for everyday knowledge work and programming tasks. Some testers believe it shows significant improvements over GPT-5.5, with overall capabilities approaching Anthropic’s Fable 5, though Anthropic still holds an advantage in writing tasks.
Other researchers have noted that Sol triggers fewer safety restrictions on life science-related issues, which may make it more appealing in biological research scenarios. Meanwhile, social media platforms have reported that GPT-5.6 may be the final generation in the 5.x series, with GPT-6 potentially already in development—though this claim has not been confirmed by OpenAI.