GPT 5.6 Luna
OpenAI's cost-optimized general-purpose language model designed for high-volume AI workloads, supporting long-context reasoning, multimodal understanding, and agent tooling for enterprise automation, assistants, content generation, and scalable applications.
- Modalities
- Chat
- Starting price
- From $0.9 / 1M out
- Context
- 1.1M context
OpenAI
README
Supported Functionality
| Item | Specification |
|---|---|
| Input | Text, Image |
| Output | Text |
| Context | 1,050,000 tokens |
| Max Output | 128,000 tokens |
| Vision | ✓ Supported |
| Function Calling | ✓ Supported |
Description
GPT-5.6 Luna is the efficiency-focused model in OpenAI's GPT-5.6 family, released on July 9, 2026 for cost-sensitive, high-volume workloads. It roughly corresponds to the nano tier in earlier GPT-5 families.
Despite its lightweight positioning, the model retains a 1,050,000-token context window, a 128,000-token maximum output, six reasoning-effort levels, image input, function calling, and Structured Outputs. As part of the GPT-5.6 family, it also supports persisted reasoning, Programmatic Tool Calling, explicit prompt caching, multi-agent orchestration in beta, and Responses API tools for retrieval, code execution, computer interaction, MCP, and Skills.
Key Capabilities
- High-Throughput Text Processing: Designed for large request volumes and cost-sensitive applications such as batch classification, summarization, rewriting, extraction, and content normalization.
- Configurable Reasoning: Supports
none,low,medium,high,xhigh, andmaxreasoning effort to balance quality, latency, and consumption by task difficulty. - Long-Context Processing: A 1,050,000-token context can hold document collections, conversation history, and extensive business material, with up to 128,000 output tokens.
- Agent and Tool Use: Function calling, Programmatic Tool Calling, MCP, tool search, and multi-agent orchestration enable multi-step automated workflows.
- Vision Understanding: Image input and
originalimage detail support screenshots, charts, scanned documents, and other visual material. - Structured Outputs: JSON Schema-constrained generation produces reliable structured results for data pipelines, business rules, and downstream APIs.
- Retrieval and Execution: Web Search, File Search, Code Interpreter, Hosted Shell, Apply Patch, and Computer Use extend knowledge access and task execution.
Technical Strengths
| Feature | Benefit |
|---|---|
| Low Per-Token Cost | Current rates of $0.20 per million input tokens and $1.20 per million output tokens support cost-sensitive deployment at scale. |
| Six Reasoning Levels | Simple tasks can use none or low to reduce latency, while harder tasks can receive more reasoning effort. |
| 1.05M-Token Context | Processes large document sets or long-running task state in one request, reducing chunking, retrieval, and assembly overhead. |
| Persisted Reasoning | Carries reasoning state through consecutive tool calls and turns, reducing repeated work and preserving continuity. |
| Programmatic Tool Calling | Allows tools to be invoked during code execution, reducing round trips between the model and application. |
| Broad Hosted Tool Support | Retrieval, code execution, shell access, computer interaction, MCP, and Skills cover the path from information gathering to action. |
Pricing
Token-based pricing
Tiered by input prompt tokens (incl. cache): once over the threshold, the whole request is billed at the higher tier.
| Token Type | Short context ≤272K | Long context >272K |
|---|---|---|
| Input | $0.15$0.2-25% | $0.3$0.4-25% |
| Output | $0.9$1.2-25% | $1.35$1.8-25% |
| Cached input | $0.015$0.02-25% | $0.03$0.04-25% |
| cache_write | $0.1875$0.25-25% | $0.375$0.5-25% |