GPT-6 Astra
OpenAI's newest flagship: an agent that works your computer for you
included with Max plan
- Context
- ~1,000,000 tokens
- Output
- up to 128,000 tokens
- Input
- Text, images
- Type
- Agentic, with reasoning
- Developer
- OpenAI
- Access
- ChatPlus PRO
GPT-6 Astra is OpenAI's newest flagship model, unveiled on September 3, 2026. The emphasis is not on chat answers but on actions: Astra can drive a browser and apps, fill out forms, understand a codebase and carry long multi-step tasks through to a result with minimal supervision. It understands text and images and holds context up to 1M tokens. On ChatPlus it is available on the Max plan.
Astra is a pivot from "smart chat" to "digital worker." OpenAI says it plainly: anything you do on a computer, the model can do for you. It looks at the screen, plans the steps, clicks, types, checks the result — and repeats until the task is solved. Hence the launch's emphasis not on pretty answers but on real work: forms, CRM, calendars, building sites, UI testing, data analysis.
On benchmarks the picture is mixed, and it's more honest without the hype. Where Astra is genuinely ahead is computer use and engineering and cybersecurity tasks (on ExploitBench it was the first to hit 100%, which led OpenAI to assign the model a "critical" level under its own safety framework for the first time). But on "general intelligence" independent measurements put it roughly on par with the previous flagship — at a noticeably higher price. This is a model for a specific class of tasks, not a universal leap over everything.
On ChatPlus, GPT-6 Astra is available on the Max plan. We connected it right after release, so you can be among the first to try OpenAI's new flagship here.
What it's good at
Working the computer
Drives a browser and apps: fills out forms, finds data, assembles documents — like a person at the screen.
Agentic tasks
Runs long multi-step scenarios, calls tools on its own and carries the work through to the end.
Programming
Understands the whole codebase, makes edits, runs tests and checks the result.
Science and analysis
Strong in math and research tasks, builds charts and parses data.
GPT-6 Astra and every other model — one plan, one window.
Where it's strongest
GPT next to the others
How GPT-6 Astra and the flagships of other labs look side by side. They're all in one ChatPlus subscription.
| Claude | DeepSeek | GPT | Gemini | Grok | GLM | Kimi | Qwen | Muse | GLM 5.3 Flash | |
|---|---|---|---|---|---|---|---|---|---|---|
| Developer | Anthropic | DeepSeek | OpenAI | xAI | Z.ai | Moonshot | Alibaba | Meta | Z.ai | |
| Model | Sonnet 4.6 | DeepSeek V4.1 Reasoner | GPT-6 Astra | Gemini 3 Pro | Grok 4 | GLM 5 | Kimi K3 | Qwen 3.8 Max | Muse Spark 1.3 | GLM 5.3 Flash |
| Context | 200K | 128K | ~1M | 1M | 256K | 1M | 256K | 256K | 1M | 1M |
| Understands images | — | — | ||||||||
| Live web data | — | — | — | — | — | — | — | |||
| Open source | — | — | — | — | — | |||||
| Key strength | Text and code | Reasoning | Computer use | Multimodality | Timeliness | Code and agents | Long context | Complex tasks | Agents and code | Code and agents |
| On ChatPlus |