OpenAI launches GPT-5.6 Sol, plus Terra and Luna
OpenAI launched GPT-5.6 Sol to general users with lower-cost Terra and Luna. Sol scored 91.9% on Terminal-Bench 2.1 in ultra mode and matched Mythos on ExploitBench using about one-third the tokens.
OpenAI released GPT-5.6 Sol to general users today and introduced two lower-cost siblings, Terra and Luna. The public launch follows a roughly two-week preview that the U.S. Department of Commerce limited to about 20 trusted partners.
OpenAI assigned names to the three models for the first time. Sol is the new flagship, Terra is positioned as an everyday option said to match GPT-5.5 at half the price, and Luna is the lowest-cost tier. OpenAI set Sol pricing 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. The company said the tiers can change independently over time.
Sol includes two new controls. A max reasoning effort setting lets the model use more compute on complex problems. An ultra mode delegates parts of multi-step tasks to subagents to handle workflows.
On Terminal-Bench 2.1, a benchmark that measures command-line and tool-driven tasks, standard Sol scored 88.8% and ultra mode reached 91.9%. Those results are higher than Anthropic’s Claude Mythos 5 at 88.0%, Claude Fable 5 at 84.3% and Claude Opus 4.8 at 78.9%. Google’s Gemini 3.1 Pro Preview registered 70.7% on the same test.
On ExploitBench, which assesses a model’s ability to find and exploit software vulnerabilities, OpenAI reported that Sol matched a restricted preview of Anthropic’s Mythos while consuming about one-third as many tokens. OpenAI said Sol does not cross the company’s internal ‘Cyber Critical’ threshold.
Early testers described performance improvements. Theo, a developer and CEO of the AI platform T3 Chat, called Sol ‘world leading in computer use’ and said it fixed issues he had with GPT-5.5 and understands subagents. Dan Shipper, whose team ran month-long internal tests, compared Sol to a high-performance daily work system and described Anthropic’s Fable as a more extreme option for high-end tasks.
Researcher Daichi Konno reported that Sol outperforms GPT-5.5 and approaches Fable 5 on many tasks. Konno noted that Sol’s safeguards did not trigger on several life-science questions during his testing.
The release arrives amid several frontier-model updates from other labs. SpaceXAI shipped Grok 4.5 yesterday and Meta released Muse Spark 1.1 earlier today. Pricing comparisons place Sol between higher-cost U.S. frontier models and lower-cost offerings from some Chinese providers. Anthropic lists Claude Fable 5 at $10/$50 per million tokens. Google’s Gemini 3.1 Pro is priced at $2/$12 per million tokens. xAI’s Grok 4.5 is $15/$75 per million tokens. Some Chinese providers offer models at substantially lower per-token rates.
A separate roadmap-tracking account on social media posted that GPT-5.6 could be the final release in the 5.x series and that GPT-6 might arrive on a larger base within about a month; that timing is presented as the account’s claim.
OpenAI made GPT-5.6 Sol, Terra and Luna available to general users today.
The material on GNcrypto is intended solely for informational use and must not be regarded as financial advice. We make every effort to keep the content accurate and current, but we cannot warrant its precision, completeness, or reliability. GNcrypto does not take responsibility for any mistakes, omissions, or financial losses resulting from reliance on this information. Any actions you take based on this content are done at your own risk. Always conduct independent research and seek guidance from a qualified specialist. For further details, please review our Terms, Privacy Policy and Disclaimers.
Articles by this author
