Key facts
- OpenAI has released its new flagship AI model, GPT-5.6 Sol, along with two smaller models, Terra and Luna.
- GPT-5.6 Sol achieved 91.9% on the Terminal-Bench 2.1 benchmark in its ultra configuration.
- Sol matched Anthropic's restricted Mythos Preview on ExploitBench, using approximately one-third of the tokens.
- Pricing for Sol is $5/$30 per million input/output tokens, positioning it between premium U.S. models and lower-cost Chinese challengers.
- Early testers have praised Sol's capabilities in computer use and its ability to handle complex tasks.
- Sol's safeguards did not trigger on life-science questions, suggesting potential utility in biology research.
OpenAI has officially launched its latest flagship AI model, GPT-5.6 Sol, making it available to the public. This release follows a two-week restricted preview managed by the U.S. Department of Commerce. The company also introduced two smaller, more affordable models: Terra, which reportedly matches GPT-5.5's performance at half the price, and Luna, a budget-friendly option.
GPT-5.6 Sol introduces new naming conventions for OpenAI, moving away from numerical designations to tiered names reflecting capability. Sol is positioned as the top-tier model, with Terra and Luna offering different price and performance points. Pricing for Sol is set at $5 per million input tokens and $30 per million output tokens, while Luna is priced at $1 and $6 respectively. The model includes new features like a 'max reasoning effort' knob for extended thinking and an 'ultra mode' that utilizes sub-agents.
Benchmark results indicate Sol's strong performance. In its ultra configuration, it achieved 91.9% on Terminal-Bench 2.1, a test evaluating command-line workflows. The standard Sol scored 88.8%. Both configurations outperformed Anthropic's Claude Fable 5 (84.3%) and Claude Opus 4.8 (78.9%), and Google's Gemini 3.1 Pro Preview (70.7%). On ExploitBench, which assesses the ability to find and weaponize software vulnerabilities, Sol matched Anthropic's restricted Mythos Preview while consuming significantly fewer tokens, and OpenAI states it remains within their 'Cyber Critical' risk framework.
Early access users have provided positive feedback. Theo, CEO of T3 Chat, described Sol as "world leading in computer use" and noted it fixed issues present in GPT-5.5, highlighting its determination and understanding of sub-agents. Dan Shipper of Every compared GPT-5.6 to a Porsche for daily knowledge work and coding, contrasting it with Anthropic's Fable, which he likened to a warp drive for more extensive tasks. Researcher Daichi Konno found Sol to be a significant improvement over GPT-5.5 and comparable to Claude Fable 5, particularly noting that Sol's safeguards did not interfere with life-science queries, potentially making it a preferred choice for biology-related work.
The release occurs in a highly competitive landscape. OpenAI's launch follows closely behind xAI's Grok 4.5 and Meta's Muse Spark 1.1. Anthropic's Fable 5 has recently transitioned from subscription plans to usage credits. Google's Gemini 3, released in November 2025, is now the oldest among the current flagship models from major U.S. AI labs. Leaks suggest that GPT-5.6 is the final model in the 5.x series, with GPT-6 anticipated within a month, and imminent releases also expected from Anthropic and DeepSeek.