OpenAI: GPT-6 Luna Pro

GPT-6 Luna Pro is the higher-compute reasoning tier of OpenAI’s GPT-6 Luna family, exposed through the gpt-6-luna API with an elevated reasoning configuration (often cataloged separately from the default medium setting). It targets focused, high-volume production workloads—routing, extraction, lightweight agents, and Codex-style coding—where teams want GPT-6-era capability at the lowest price point in the lineup. Like base Luna, it supports very long context (about 1.05M tokens), large outputs (up to 128K tokens), text and image inputs, tool use, structured outputs, prompt caching, and adjustable reasoning effort from none through max.

OpenAI released GPT-6 Luna alongside GPT-6 Sol on September 22, 2026, distilling training and alignment methods from GPT-6 Astra into faster, cheaper models. Official evaluations emphasize agentic professional work (AutomationBench, Agents’ Last Exam), coding (FrontierCode, DeepSWE), and computer use (OSWorld 2.0), with Luna positioned as the cost-efficiency leader on the GPT-6 cost–intelligence curve. API pricing for Luna is $0.10 per million input tokens and $0.50 per million output tokens (with discounted cached reads), roughly half the promotional rates of GPT-5.6 Luna.

Independent leaderboards at max (or equivalent top) effort report strong general and STEM scores—notably GPQA-Diamond near 90%, MMLU-Pro in the mid-80s, and competitive software-engineering agent results on DeepSWE and SWE-bench Verified—while agent-heavy office automation benchmarks remain modest relative to larger Sol or frontier models. GPT-6 Luna Pro is proprietary and available only via OpenAI’s API and partner surfaces such as ChatGPT Work, Codex, and GitHub Copilot, not as downloadable weights.