GPT-6 Luna (Non-reasoning)

GPT 6 Luna is an efficiency-focused large language model developed by OpenAI, released on September 22, 2026. It serves as the most lightweight and cost-effective entry in the GPT-6 generation, succeeding the GPT-5.6 Luna model. The model is specifically designed for high-volume, latency-sensitive workloads and recurring tasks that require advanced intelligence without the higher operational costs of the flagship Astra or mid-tier Sol models.

The model features an expanded context window of 1,050,000 tokens, enabling the analysis of extensive documents and multi-file codebases in a single request. It is a multimodal system that accepts text and image inputs and provides text-based outputs. A core feature of the model is its support for granular reasoning effort controls, which allow users to adjust the depth of the model's processing between various tiers to optimize the balance between speed and quality for specific task requirements.

OpenAI reports that GPT 6 Luna incorporates a more natural and refined communication style, characterized by increased technical clarity and a reduction in jargon compared to previous generations. In alignment evaluations, it has shown improved reliability in technical and coding conversations, specifically reducing the rate of misleading claims about its own work. The model is also optimized for computer-use capabilities, allowing it to navigate software interfaces and execute tool-enabled workflows with higher precision.

Designed for agentic applications, GPT 6 Luna is frequently utilized for information extraction, large-scale summarization, and developer assistance. Its architecture benefits from optimizations in prompt caching and inference efficiency, enabling high throughput for enterprise-scale applications that demand frequent and repeatable model interactions.