Core capabilities
- 1.05M context with up to 128K output
- Image input, function calling and structured outputs
- Reasoning levels from none through max
- Search, code, shell, computer-use and MCP tools
OpenAI / current
The cost-sensitive GPT-5.6 variant for high-volume, decomposable production tasks.
Best for classification, extraction, routing and subagent work that can be evaluated at scale. Luna is the high-volume member of GPT-5.6, yet retains the family's long context, 128K output, image input, reasoning controls and tool integrations.
Use it for extraction, classification, routing, bulk transformation and bounded subagents. Add schema validation, confidence gates and an automatic Terra/Sol fallback for ambiguous or high-impact cases.
Efficient models need crisp task boundaries and an escalation path for low-confidence results. A large context window is a capacity limit, not a recommendation to send every document. Retrieval, prompt caching and escalation thresholds usually matter more for throughput.
Specifications and positioning checked against provider documentation.
Open official source↗