OpenAI’s supplied API and product sources support GPT-5.6 Luna as a real API model positioned for cost-sensitive, high-volume use, with text and image input, text output, Responses API tools, a 1,050,000-token context window, and 128,000 max output tokens. Pricing is unusually clear in the supplied evidence: $1.00/M input, $0.10/M cached input, and $6.00/M output, with Artificial Analysis independently repeating the same price and 1.0M context figure. Coding evidence is positive but incomplete. DeepSWE/DataCurve lists gpt-5.6-luna[max] at 67%±4% over 113 tasks, with reported average cost, output-token, and step counts, suggesting capable agentic coding at moderate cost. LiveCodeBench direct search found no GPT-5.6/Luna listing, and Requesty marks LiveCodeBench as N/A while giving a separate Coding Index. No SWE-bench result was found. Terminal-Bench/Aider were checked, but the supplied sources do not provide a model-specific score. Public preference and general intelligence signals are mixed. Artificial Analysis reports Intelligence Index 46, high throughput at 202.6 output tokens/s, but TTFT latency of 7.83s. LMArena dataset evidence lists GPT 5.6 Luna xHigh at rank 18, while Prolific’s HUMAINE report places Luna #36 of 54. Overall, Luna looks like a strong low-cost OpenAI production option, but not a top frontier model; confidence is reduced by missing SWE-bench, LiveCodeBench, and Terminal-Bench/Aider results.
