GPT-5.6 Luna: Complete Specifications, Pricing, API Access & Use Cases (2026)

What Is GPT-5.6 Luna?

GPT-5.6 Luna is OpenAI’s cost-focused multimodal reasoning model, released within the GPT-5.6 family in 2026, with a 1,050,000-token context window, text-and-image input, text output, and API pricing of \$1 per million input tokens and \$6 per million output tokens as of July 2026.

OpenAI describes Luna as a GPT-5.6 model optimized for cost-sensitive, high-volume workloads. It is the fastest and most cost-efficient model in the GPT-5.6 family and broadly corresponds to the nano tier used in earlier GPT-5 generations.

The model is intended for applications where request volume, latency, and token cost are important, but meaningful reasoning capability is still required. Typical search intent around GPT-5.6 Luna includes its context window, token pricing, image support, API availability, agent features, and differences from GPT-5.6 Terra or earlier nano-class models.

GPT-5.6 Luna is available through the OpenAI API, Codex, and eligible Work in ChatGPT environments. It is not independently selectable in standard ChatGPT conversations as of July 2026.

What Are GPT-5.6 Luna’s Key Specifications and Pricing?

The following specifications reflect official OpenAI documentation and the Gate.AI model-card and API documentation available as of July 2026.

OpenAI’s model page verifies the 1.05-million-token context window, 128,000-token maximum output, February 2026 knowledge cutoff, supported modalities, token prices, long-context multiplier, API features, tools, and tier-based limits.

Prompt caching can materially reduce repeated-input costs. OpenAI lists cached input at \$0.10 per million tokens and cache writes at 1.25 times the uncached input rate. Explicit cache breakpoints and a minimum 30-minute cache life are documented for the GPT-5.6 family.

What Can GPT-5.6 Luna Do That Makes It Useful in Production?

Process high-volume conversational requests

GPT-5.6 Luna may fit customer-support triage, internal chat assistants, product help systems, intent detection, and routine response generation. Its \$1 input and \$6 output pricing can make it more practical than larger GPT-5.6 tiers for applications processing many requests.

Lower cost does not guarantee lower total spend. Long outputs, repeated tool calls, and prompts exceeding 272,000 input tokens can materially increase the cost of each workflow.

Analyze large document collections

The 1.05-million-token context window allows applications to submit long reports, transcripts, policy collections, source-code repositories, or extended conversation histories in a single request.

This can simplify some document workflows, but a large context window should not automatically replace retrieval. Selective retrieval can improve relevance, reduce latency, and avoid the higher rates applied to requests above the 272,000-token threshold.

Generate structured classifications and extractions

Luna supports structured outputs and function calling, making it relevant for tagging, metadata generation, entity extraction, ticket routing, form processing, and machine-readable content transformation.

Production teams should validate schemas, handle refusal or truncation cases, and test extraction accuracy on representative documents rather than assuming consistent performance across every format.

Support lightweight agent workflows

Through the Responses API, GPT-5.6 Luna supports tools including web search, file search, code interpreter, hosted shell, apply patch, computer use, MCP, and tool search.

These capabilities can support routine agent loops such as retrieving information, processing files, calling internal functions, applying code changes, or completing multi-step operational tasks. Tool permissions, network access, approval steps, and output validation remain important because an AI-generated action can still be incomplete or incorrect.

Interpret images and return text

GPT-5.6 Luna accepts image input, allowing it to analyze screenshots, diagrams, interfaces, scanned documents, forms, charts, and other visual content. The model returns text rather than native image, audio, or video output.

Image quality, small text, complex layouts, and ambiguous visual context can affect the reliability of the result.

What Are GPT-5.6 Luna’s Supported Modalities?

OpenAI identifies text as an input-and-output modality, image as input only, and audio and video as unsupported native modalities.

Tool availability should not be confused with native modality support. For example, the Responses API may allow Luna to call an image-generation tool, but this does not make native image output part of the Luna model itself.

Where Does GPT-5.6 Luna Fall Short?

GPT-5.6 Luna is optimized for cost-sensitive workloads rather than the highest capability available in the GPT-5.6 family. More demanding research, complex software engineering, extended planning, advanced scientific analysis, or difficult agent tasks may justify evaluating GPT-5.6 Terra or Sol.

Its large context window also creates a cost trade-off. Inputs above 272,000 tokens are charged at twice the normal input rate, while output is charged at 1.5 times the standard rate for the entire request. Applications that indiscriminately attach large document collections may therefore spend more than expected.

The documented knowledge cutoff is February 16, 2026. Information after that date should be supplied through retrieved sources, tools, or current databases rather than assumed to be present in the model’s internal knowledge.

Fine-tuning is not supported as of July 2026. Teams requiring task-specific adaptation must use prompting, retrieval-augmented generation, tools, workflow controls, or a different model with supported customization.

GPT-5.6 Luna may still produce unsupported statements, overlook constraints, misread an image, return an incomplete structured response, or choose an unsuitable tool. This is a general AI limitation and is not unique to Luna.

Outputs involving medicine, law, finance, employment, public safety, cybersecurity, or other high-impact decisions require qualified human review. The model should assist research and workflow execution rather than act as the sole decision-maker.

What Is GPT-5.6 Luna Best Used For?

The phrase "best used for" is scenario-dependent. GPT-5.6 Luna may be a suitable choice when throughput, latency, long context, and lower GPT-5.6 pricing matter more than obtaining the maximum capability available in the family.

Teams evaluating economical alternatives may also compare GPT-4o mini, Gemini 2.5 Flash, and Claude Haiku 4.5. These models differ in pricing, tool ecosystems, context handling, modality support, and provider-specific API behavior.

How Does GPT-5.6 Luna Compare to GPT-5.6 Terra and GPT-5.4 Nano?

The comparison does not identify a universal winner. Luna may fit teams seeking GPT-5.6 capabilities at a lower price than Terra. Terra may be more appropriate when a workload benefits from additional capability and can justify the higher token cost. GPT-5.4 Nano may remain relevant for simpler tasks where minimizing input cost is the dominant requirement.

OpenAI’s Luna model page presents GPT-5.6 Terra and GPT-5.4 Nano as nearby pricing comparisons, making them useful reference points for cost and tier positioning.

How Do I Access GPT-5.6 Luna Through Gate.AI?

According to the Gate.AI model-card, GPT-5.6 Luna is available under the model ID openai/gpt-5.6-luna. Gate.AI provides an OpenAI-compatible API that uses the base URL https://api.gate.ai/openai/v1, bearer-token authentication, and model identifiers in provider/model-name format.

Gate.AI documentation lists support for OpenAI-compatible integrations, API-key management, fixed-model selection, model discovery, and automatic routing when enabled in the Gate.AI console. The applicable model price and account availability should be confirmed on the current Gate.AI model-card or account console before production deployment.

Python Example

import os
from openai import OpenAI
client = OpenAI(
    api_key=os.environ["GATEAI_API_KEY"],
    base_url="https://api.gate.ai/openai/v1",
)
response = client.chat.completions.create(
    model="openai/gpt-5.6-luna",
    messages=[
        {
            "role": "user",
            "content": "Classify this support request and explain the label briefly."
        }
    ],
)
print(response.choices[0].message.content)

curl Example

curl https://api.gate.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer $GATEAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.6-luna",
    "messages": [
      {
        "role": "user",
        "content": "Classify this support request and explain the label briefly."
      }
    ]
  }'

Gate.AI documents POST /chat/completions, streaming support, GET /models, bearer authentication, and the /openai/v1 base path for its OpenAI-compatible API.

Developers using OpenAI directly can access the model with the provider model ID gpt-5.6-luna through the Responses API or Chat Completions API. OpenAI also documents streaming, function calling, structured outputs, Batch API support, and multiple Responses API tools.

FAQs

What is GPT-5.6 Luna’s context window?

GPT-5.6 Luna has a 1,050,000-token context window and supports outputs of up to 128,000 tokens as of July 2026. OpenAI applies higher pricing to the entire request when input exceeds 272,000 tokens.

How much does GPT-5.6 Luna cost?

OpenAI prices GPT-5.6 Luna at \$1 per million input tokens, \$0.10 per million cached input tokens, and \$6 per million output tokens as of July 2026. Cache writes cost \$1.25 per million tokens.

How can developers access GPT-5.6 Luna?

Developers can access it through OpenAI using gpt-5.6-luna. According to the Gate.AI model-card, the Gate.AI model ID is openai/gpt-5.6-luna, used with Gate.AI’s OpenAI-compatible API and a Gate.AI API key.

What workloads may suit GPT-5.6 Luna?

GPT-5.6 Luna may suit high-volume chat, tagging, extraction, document analysis, retrieval workflows, image-aware text tasks, and lightweight agents. More difficult research, engineering, or long-horizon agent tasks may justify testing GPT-5.6 Terra or Sol.