OpenAI has begun a limited preview of its new GPT-5.6 model series, according to an announcement surfaced by Techmeme. The lineup includes three tiers: Sol, billed as the flagship model; Terra, described as a balanced model for everyday work; and Luna, a faster option.
The company is positioning Sol at the high end of the range. According to OpenAI's announcement via Techmeme, GPT-5.6 Sol matches a model called Mythos Preview on a benchmark named ExploitBench. The new release also introduces an "Ultra mode" that uses subagents to handle complex workflows, along with a "max reasoning" setting aimed at deep problem-solving.
MacRumors, cited by Google News, likewise reports that OpenAI is launching Sol, Terra, and Luna in limited preview.
The three-model structure is notable because it signals a tiered approach: a top-capability model for the hardest tasks, a general-purpose model for routine use, and a speed-focused model for lighter work. The mention of subagents in Ultra mode points to systems that can break large, multi-step tasks into smaller pieces handled by coordinated helper processes — an increasingly common design pattern for AI tools meant to tackle involved workflows rather than single questions.
The sources provided do not specify pricing, exact availability dates, how broad the limited preview will be, or detailed performance figures beyond the ExploitBench comparison. It is also not clear from these items what Mythos Preview is or how ExploitBench scores are measured.
Why it matters: a tiered model lineup with subagent capabilities suggests AI providers are increasingly building tools designed to carry out complex, multi-step work — not just answer prompts — which could reshape how businesses and everyday users rely on these systems.