OpenAI: GPT Luna Latest

GPT Luna Latest refers to GPT-6 Luna, OpenAI's most efficient tier in the GPT-6 family for focused, high-volume work. Released in September 2026 alongside GPT-6 Sol, it extends the cost-efficiency advances of GPT-6 Astra into a smaller, faster model aimed at agents, coding helpers, and everyday professional tasks where price and latency matter more than peak frontier depth. It is available through the OpenAI API as gpt-6-luna, in ChatGPT Work and Codex for paid tiers, and as the GPT-6 option for Free and Go users in the ChatGPT desktop app.

The model supports text and image inputs, text outputs, and configurable reasoning effort from none through max, with a 1.05 million-token context window and up to 128,000 output tokens. API pricing is $0.10 per million input tokens and $0.50 per million output tokens, roughly half the promotional rates of GPT-5.6 Luna. OpenAI positions Luna for scalable automation, multi-step tool use, and coding agents when Sol or Astra would be unnecessarily expensive.

On OpenAI's own evaluations at max effort, GPT-6 Luna reaches 66.6% on DeepSWE v1.1, 50.9% on Agents' Last Exam, 20.7% on AutomationBench, and 52.7% on OSWorld 2.0 offline, with substantially lower factual error rates at higher reasoning settings. Independent harnesses report strong knowledge and multimodal scores, including about 86% on MMLU-Pro and 90.5% on GPQA Diamond at max effort, while Artificial Analysis places its Intelligence Index near GPT-5.6 Luna, highlighting that the main upgrade is cost performance rather than a large jump in raw capability.