OpenAI: GPT-6 Luna (batch)

GPT-6 Luna (batch) is the batch-inference SKU of OpenAI's GPT-6 Luna (openai/gpt-6-luna:batch), aimed at focused, high-volume workloads such as classification, lightweight agents, and cost-sensitive coding assistance. It shares the same weights and capabilities as the standard gpt-6-luna endpoint but is invoked through the Batch API (v1/batch), which trades latency for roughly half the per-token cost of synchronous calls—making it practical for offline enrichment, evaluation pipelines, and large backlogs of similar requests.

The model supports text and image inputs, configurable reasoning effort, tool use, and a very large context window (about 1.05M tokens with up to 128K output tokens). OpenAI positions Luna below GPT-6 Sol and GPT-6 Astra on the cost–intelligence curve while still delivering strong agentic performance on tasks such as DeepSWE software engineering, Terminal-Bench terminal workflows, and professional automation benchmarks when run at higher effort settings.

Reported evaluations use the same GPT-6 Luna (max) configurations as the live API; batch jobs do not change model behavior, only scheduling and pricing. Developers typically pair Luna with prompt caching and effort tuning to balance quality against throughput on repetitive or long-context batch jobs.