GPT-6 Luna by OpenAI (Text + Vision)

Overview

GPT-6 Luna is a GPT‑6 family reasoning model made by OpenAI. It takes text prompts and can also read images, then returns a text answer. It’s built for focused, high-volume work where you still need strong reasoning and reliable formatting. The long context window helps when you must keep large conversation state or long documents in one run.

What you can build

• Customer support copilots that keep long conversation history without losing context
• Coding helpers that review diffs, explain errors, and propose fixes in plain language
• Screenshot and UI review bots that explain what’s on screen and what to change
• Document Q&A workflows where users paste large passages and ask targeted questions
• Data labeling assistants that apply consistent rules and return structured text
• Tool-using agents that decide when to call functions and how to use results

Inputs

• A written instruction in plain text that describes what you want the model to do.
• Optional conversation history as a list of messages (for multi-turn chat flows).
• Optional images for the model to analyze, provided as PNG, JPEG/JPG, WebP, or non-animated GIF files. Wiro runs for this model support up to 10 images per request.
• Optional system-level instructions that set global rules for tone, style, and constraints. Keep this within 5,000 characters.
• An optional reasoning level setting that controls how much deliberate reasoning the model applies.
• An optional verbosity control that influences how short or detailed the answer is.
• An optional maximum-length limit for the response. This model supports up to 128,000 output tokens.
• Optional tool definitions (functions) when you want the model to request tool calls.
• Optional tool behavior controls that let you force tool use, forbid tool use, or let the model decide.
• Optional chat history identifiers (a user ID and session ID) when you want Wiro to store and reuse prior turns.
• Optional continuation inputs for tool workflows, where you resume a prior run by providing the prior run token plus the tool results.

Outputs

GPT-6 Luna returns a structured response with a primary text answer. In tool-enabled runs, it can also return one or more tool-call requests that your app can execute, followed by a continued text response after you supply tool results. The response also includes completion status and token-usage accounting, which helps you track how much context and output the run consumed.

Recommended settings

• Use the Responses-style workflow when you need built-in tools and function calling.
• If you run in a Chat Completions-style workflow and you need function calling, keep reasoning turned off for that turn.

Limitations

• GPT-6 Luna accepts text and images as input, but it produces text output only.
• The model’s knowledge is current up to May 18, 2026, so newer events may be missing.
• Image understanding depends on image clarity. Blurry screenshots, tiny text, and heavy compression reduce accuracy.
• Scanned documents and low-contrast photos can cause missed details or incorrect OCR-like reads.
• Very large requests may hit platform payload limits. Vision-capable requests support up to 512 MB total payload, and up to 1,500 images per request in the underlying OpenAI vision input rules.

Safety & compliance

OpenAI applies safety training and monitoring across the GPT-6 family. The model may refuse disallowed requests, including content involving sexual content with minors, instructions for wrongdoing, or other policy-restricted content. If you build tool-using workflows, add your own confirmation steps for actions that could cause real-world harm, like sending messages, making purchases, or running destructive code.