Introducing GPT-6 Sol and Luna

2026-09-22 · OpenAI

Introducing GPT-6 Sol and Luna

Overview

The original announcement introduces GPT-6 Sol and Luna, described as two models that bring frontier intelligence to everyday work. According to the available text, they offer different balances of capability and cost. The announcement does not provide additional technical, commercial, or operational details.

The two models

  • Names: GPT-6 Sol and GPT-6 Luna.
  • Shared positioning: bringing frontier intelligence to everyday work.
  • Difference: different balances of capability and cost.

Capability and cost

The phrase “different balances of capability and cost” is the central description of how the two models differ. It indicates that Sol and Luna are not identical options. However, the original text does not say which model emphasizes capability, which emphasizes cost, or how the trade-off is measured. It also does not provide pricing, performance figures, benchmarks, or target users. Therefore, the only confirmed point is that the two models are positioned differently along a capability-cost spectrum.

Everyday work

The announcement says the models are intended for everyday work. This framing suggests a focus on practical, routine, or regular tasks rather than only experimental or specialized use. Still, the original text does not list specific tasks, product interfaces, API availability, deployment methods, or whether the models are aimed at individuals, teams, or enterprises. Those details are absent.

Frontier intelligence

The models are said to bring frontier intelligence. This is a broad claim about capability level, but the announcement does not define what “frontier” means in this context. It does not mention model size, architecture, training data, context length, modalities, reasoning abilities, or evaluation results. As a result, the phrase should be read as the announcement’s positioning rather than as a verified specification.

What is known

  • GPT-6 Sol and Luna are being introduced.
  • Both are associated with frontier intelligence.
  • Both are aimed at everyday work.
  • They have different balances of capability and cost.

What is not known

  • Model size, architecture, or training details.
  • Context window, modalities, or reasoning features.
  • Pricing, rate limits, or availability.
  • Benchmarks, comparisons, or release dates.
  • Which model is more capable and which is more cost-oriented.

Why the framing matters

The key idea in the original text is choice. By presenting two models with different capability-cost balances, the announcement suggests that users may be able to select an option based on their needs. The text does not explain how that selection should be made, what the trade-offs are in practice, or what use cases fit each model. Any such guidance would go beyond the source.

Summary

The available announcement is brief. It confirms the introduction of GPT-6 Sol and Luna, their shared goal of bringing frontier intelligence to everyday work, and their different balances of capability and cost. It does not provide specifications, pricing, availability, benchmarks, or release information. A full assessment would require additional official details.

Source