<MCP.Playground/>
online

GPT-6 Luna

Runs MCP tools

OpenAI’s fast GPT-6 tier, released 22 September 2026. OpenAI says it matches last generation’s Sol. Checking that on your own MCP tools takes a minute.

Vendor
OpenAI
Released
22 September 2026
Context
1.05M tokens
Input
Text, image, file

Paste a server URL, pick a model, and watch it call your tools in a real conversation. No install.

No server of your own? Leave it blank and use one of the public mock servers.

MCP Evals

Can GPT-6 Luna actually complete tasks with your server?

One chat shows the tool calls work. An eval writes real tasks from your tool schemas, has GPT-6 Luna drive each one, and reports where it got the wrong answer even though every call succeeded. Pick GPT-6 Luna as the driver.

Run evals →See a sample

What GPT-6 Luna is

GPT-6 Luna is the fast tier of the GPT-6 series, below Sol. It shares the 1.05M window of the rest of the series. OpenAI aims it at high-volume routine work such as summarisation, extraction, classification and lightweight agent tasks. Luna Pro is the same model with reasoning mode set to pro.

What OpenAI says

  • DeepSWE v1.1: 66.6%, against 68.8% for GPT-6 Sol.
  • Matches GPT-5.6 Sol, by OpenAI’s account.

These are the vendor’s own claims, not measurements of ours. Run the model against your own server to find out whether they hold for your tools.

How it reaches MCP

  • The OpenAI Responses API with remote MCP servers as a tool type. Model id gpt-6-luna.
  • Any MCP client over OpenRouter as openai/gpt-6-luna or openai/gpt-6-luna-pro, which is what runs here

What to watch for

  • The claim to check is the one against GPT-5.6 Sol, and both are on this site. Run them on the same prompt. If Luna holds up on your tools, you can use the faster model. If not, you have found where your schemas need a stronger model.
  • “Lightweight agentic” is OpenAI’s own scoping. Expect the gap to Sol to show on long, multi-step tool chains before it shows on single calls.
  • Like the rest of GPT-6, it takes seed but not temperature on OpenRouter, so use seed when you need repeatable tool calls.

Run GPT-6 Luna and GPT-5.6 Sol on the same complex-schema prompt. analyze_data and configure_workflow are the two tools where a small model is most likely to produce arguments that are close but wrong. All the mock servers →

Variants available here

ModelModel ID
GPT-6 Lunaopenai/gpt-6-luna
GPT-6 Luna Proopenai/gpt-6-luna-pro

Go deeper

Frequently asked questions

Is GPT-6 Luna good enough for MCP tool calling?
For flat, well-described tools it is likely to be. OpenAI scopes it to lightweight agent work, and its DeepSWE score is close to Sol’s. Long multi-step chains are where to check it. Running it against your own server here settles the question for your tools.
What is the difference between GPT-6 Luna and Luna Pro?
Same model. Luna Pro runs with reasoning mode set to pro, which trades speed for more reasoning on each answer.
GPT-6 Luna or GPT-6 Sol?
Sol is the stronger model and scores 68.8% on DeepSWE v1.1 against Luna’s 66.6%. For high-volume, simple tool calls, Luna. For multi-step agent work where a wrong call is expensive, test Sol.
Can I use GPT-6 Luna for free?
In ChatGPT, Luna is the desktop-app model for Free and Go users. Here you can run it against any MCP server with no OpenAI key.

Sources

GPT-6 Luna for MCP: Specs & Tool Calling — Test Free | MCP Playground