<MCP.Playground/>
online

GPT-6 Astra

Runs MCP toolsReasoning

The flagship of OpenAI’s GPT-6 series, released in early September 2026, ahead of Sol and Luna. OpenAI built it for long, end-to-end work. The question for most MCP servers is whether that work needs it.

Vendor
OpenAI
Released
4 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.

In your browser

MCP Agent Studio

Server has a public URL? Chat with GPT-6 Astra on it, watch every tool call, and compare models side by side.

Open MCP Agent Studio →

On your computer

MCP Playground Desktop

Server runs on localhost or over stdio? The browser can’t reach it. The desktop app can, with GPT-6 Astra in Agent Studio. Download free

Mac (Apple Silicon)Mac (Intel)Windowsor Microsoft Store

MCP Evals

Can GPT-6 Astra 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 Astra drive each one, and reports where it got the wrong answer even though every call succeeded. Pick GPT-6 Astra as the driver.

Run evals →See a sample

What GPT-6 Astra is

GPT-6 Astra is the top of the GPT-6 series, above Sol and Luna. OpenAI announced it on 3 September 2026 and rolled it out to a limited set of organisations first, then to ChatGPT Plus, Pro, Business and Enterprise, the API and AWS. It has a 1.05M-token window and takes text, image and file input. OpenAI frames it around computer use, professional work and cybersecurity.

What OpenAI says

  • OSWorld 2.0: 72.6%, against 65.7% for GPT-5.6 Sol and 70.2% for Claude Opus 5.
  • Terminal-Bench 4.0: 57.7%, against 55.8% for Claude Fable 5.1 and 37.3% for GPT-5.6 Sol.
  • FrontierMath Tier 4: 97.6%, against 87.8% for Claude Fable 5.1.

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, which can call remote MCP servers as a tool type. Model id gpt-6-astra.
  • ChatGPT on the Plus, Pro, Business and Enterprise plans, where MCP servers connect as connectors
  • MCP Playground’s Agent Studio: paste your server URL above and run Astra against it in the browser, with no OpenAI key

What to watch for

  • Astra is the heaviest GPT-6 and runs are slower for it. OpenAI tunes Sol for recurring agent work, which is what most MCP workloads are. Test Sol on the same prompt first and move up only if it misses.
  • Its Terminal-Bench 4.0 lead was short-lived. Claude Opus 5.5, released later the same month, reports 66.4% against Astra’s 57.7%. Run both on your tools before treating Astra as the ceiling.
  • Like the rest of GPT-6, it takes seed but not temperature on OpenRouter. Use seed when you need repeatable tool calls.

Give Astra and GPT-6 Sol the same four-step task on the complex-schema mock: create_user_profile, then process_order for that user, then analyze_data on the result, then configure_workflow. If both finish with the same calls, Sol is enough for your server. All the mock servers →

Available here

ModelModel ID
GPT-6 Astraopenai/gpt-6-astra

Go deeper

Frequently asked questions

When was GPT-6 Astra released?
OpenAI announced it on 3 September 2026, starting with a limited set of organisations and widening to ChatGPT paid plans, the API and AWS over the following days.
GPT-6 Astra or GPT-6 Sol for MCP tools?
Start with Sol. OpenAI aims Astra at the hardest end-to-end work and Sol at recurring coding and agent work, which covers most MCP servers. Run the same prompt on both here. If the tool calls match, the lighter model is the one to ship.
Does GPT-6 Astra support MCP?
Yes. OpenAI’s Responses API can call remote MCP servers directly, ChatGPT supports them as connectors on paid plans, and any MCP client can drive it through function calling. You can run it against your own server here without an OpenAI key.
Is GPT-6 Astra better than Claude Opus 5.5?
On the vendors’ own numbers it depends on the benchmark: Astra leads on FrontierMath and computer use, Opus 5.5 on Terminal-Bench 4.0. Neither tells you how either handles your tool schemas. Running both against your server does.

Sources