What Is GPT-6 Astra? OpenAI's New Flagship, and How to Actually Get It

TL;DR: GPT-6 Astra is OpenAI's frontier reasoning model, generally available since September 4, 2026. It ties Claude Fable 5.1 at 53 on the Artificial Analysis Intelligence Index v4.3, pitched at coding and computer work. The API costs $10 / $50 per million tokens. The catch: ChatGPT Plus gets it only in Work and Codex, not regular Chat.

What is GPT-6 Astra?

GPT-6 Astra is OpenAI's newest flagship. It went into limited preview on September 3, 2026, first to vetted customers in OpenAI's cybersecurity program, and became generally available on September 4 (OpenAI's announcement, CNBC). It ships in ChatGPT, Codex, the OpenAI API and Amazon Web Services.

OpenAI's framing, not ours: the company calls it "the most intelligent and aligned model in the world" and claims state of the art for computer use, browsing, software engineering, cybersecurity, science and professional work. Per CNBC, OpenAI says it can fill out online forms, update customer records and create documents, spreadsheets and presentations. Greg Brockman called it a "generational leap" that could eventually be seen as AGI (Axios). The announcement names benchmarks but the community post lists no scores, so for numbers we go elsewhere.

The specs, from OpenAI's model docs:

• Context window: 1,050,000 tokens; max output: 128,000 tokens
• Input and output: text and image in, text out
• Knowledge cutoff: April 30, 2026
• Reasoning effort: low, medium, high, xhigh, max (the default is not documented)

Two things from the reception matter. Per Wikipedia's summary, Astra uses a "recurrent depth" reasoning technique that obscures some or all of its reasoning, raising monitorability concerns among safety experts: reasoning you can't read is reasoning you can't audit. And its advanced cybersecurity capabilities were initially restricted to testers. It was also reportedly trained on more than 100,000 GPUs.

The numbers, from a neutral scoreboard

These are independent measurements from Artificial Analysis on Intelligence Index v4.3, published September 7, 2026.

Model (effort)	Intelligence Index v4.3	Rank
GPT-6 Astra (max)	53	#3 of 200
Claude Fable 5.1 (max, with fallback)	53	-
GPT-6 Astra (high)	51	#6
GPT-5.6 Sol (max)	47	#14

• It got to 53 cheaply. Astra and Fable 5.1 share a $10 / $50 list price, yet at the same score Astra used about 27k output tokens per task against Fable's 78k: $3.26 per task versus $7.63.
• Coding agents are a tie. Astra scores 62 on AA's Coding Agent Index, level with Fable 5.1.
• Fewer hallucinations than Sol. At max effort, the hallucination rate falls from GPT-5.6 Sol's 92% to 51%. Better, not solved.

Why you might have seen "66" for Fable 5.1. That's from Artificial Analysis' September 1 launch write-up, which predates v4.3. The new version upgraded Terminal-Bench to v4.0 and added AutomationBench-AA, and on it both models score 53. Don't compare scores across index versions.

Mind the effort setting. These scores were measured at max or high effort, and max beats high by two points. TulexAI runs Astra at OpenAI's default effort, so don't expect these exact numbers in everyday chat. Full benchmark depth is in our Astra vs Fable 5.1 vs Sol showdown; the spec sheet lives on the GPT-6 Astra model page.

How to actually get GPT-6 Astra

This is where people get stuck. A thread on OpenAI's community forum asks for clarification because Astra was understood as announced for all ChatGPT Plus users, while Plus access is limited to Work and Codex. Here's the real breakdown, per the OpenAI Help Center (GPT-5.6 in ChatGPT, managing usage with Astra) and Notebookcheck:

Where	Price	What you get	Limits and catches
ChatGPT Plus	$20/mo	Astra in Work and Codex only. Regular Chat runs GPT-5.6 Sol	Limited Work/Codex allowance, optional credits
ChatGPT Pro	$100/mo	Astra in Chat (as "GPT-6 Pro"), Work and Codex	50 GPT-6 Pro messages a week in Chat (reported)
ChatGPT Pro	$200/mo	Same surfaces	200 GPT-6 Pro messages a week in Chat (reported)
ChatGPT Business, Enterprise	-	Astra on all surfaces	-
OpenAI API	$10 / $50 per 1M tokens	Every effort level, up to 128K output	Function calling via the Responses API only
Amazon Web Services	-	Available, per CNBC	-
Microsoft Foundry (Azure)	-	Generally available, per Microsoft	-
TulexAI	Pro $23/mo (also VIP, Elite)	Astra in chat, text and image input	25 PT per 750 tokens of prompt plus reply; no tools; 8,192-token replies

Pro prices and weekly counts are Notebookcheck's reporting; plan surfaces come from OpenAI's Help Center. Two details trip people up:

• "GPT-6 Pro" is Astra. In regular Chat on Pro, the model is called GPT-6 Pro, powered by Astra, with its own message limits. Searching the Chat picker for "Astra" is the wrong search.
• Plus access runs out faster. OpenAI's help article warns that "Astra can use your allowance faster than GPT-5.6 Sol."

We enabled Astra on TulexAI on September 8, after live checks passed for text, streaming and image input. It's on Pro, VIP and Elite, not Basic, and a tool request is refused with a clear error.

The trade-offs nobody headlines

• No function calling on Chat Completions. OpenAI's reasoning guide is blunt: "Chat Completions does not support function calling with GPT-6 Astra." Tools are Responses API only, awkward for a model pitched on computer use. More in Can GPT-6 Astra run agents?. On TulexAI, use a tool-capable model like Claude Fable 5.1 or GPT-5.6 Sol and start with our agents docs.
• Slow first token. Artificial Analysis measured 334 seconds to first token at max effort (thinking included), then 58.1 tokens per second. At high effort, 37 seconds. Fine for a hard analysis, painful for a quick question.
• Pricier than Sol. On the API, Astra is $10 / $50 per million tokens against GPT-5.6 Sol's standard $5 / $30, and Sol is on a promotional $4 / $20 at least through November 21, 2026. Sol scores 47 to Astra's 53, so ask whether six Index points are worth the premium for your workload.

Astra vs Fable 5.1 vs Sol: who should pick which?

• Pick GPT-6 Astra when the question is hard, you don't need tools, and accuracy matters more than speed.
• Pick Claude Fable 5.1 when you want the same Index score plus tool calling. It supports tools on TulexAI at the same 25 PT per 750 tokens as Astra, Pro and up, and Claude Pro ($20/mo) includes it with a reduced share of weekly limits.
• Pick GPT-5.6 Sol when cost and speed beat the last six points. Its standard API price is $5/$30 per million tokens against Astra's $10/$50, it is the standard Chat model on ChatGPT Plus, and it supports tools on TulexAI.

Two-way breakdowns: Claude Fable 5.1 vs GPT-6 Astra and GPT-5.6 Sol vs GPT-6 Astra.

What GPT-6 Astra costs in practice

On the API (OpenAI pricing), cached input is $1 per million, cache writes $12.50, and prompts over 272K tokens bill at $20 / $75. Batch and Flex are half price; Fast Mode is double. A short exchange of 250 input and 500 output tokens costs $0.0025 + $0.025 = $0.0275 at list price. Real tasks run longer: at max effort, Artificial Analysis measured $3.26 per Index task.

On ChatGPT, Astra in regular Chat starts at Pro, $100/mo, for a reported 50 GPT-6 Pro messages a week.

On TulexAI, it's 25 PT per 750 tokens of prompt plus reply on Pro at $23/mo, with Astra's hidden reasoning counted in and a 25 PT minimum per request, so a long or reasoning-heavy answer costs a multiple of 25 PT. That is the same PT price as Fable 5.1, on one subscription that also covers Claude, Gemini and image, video and voice models. The honest trade: default effort, no tools, 8,192-token replies. Need max effort, 128K outputs or tool calling? Use the API. Want Astra in a chat window without a $100 plan? That's us. See the pricing page or our ChatGPT alternative breakdown.

Frequently Asked Questions

Is GPT-6 Astra available on ChatGPT Plus?

Partly. Per OpenAI's Help Center, Plus ($20/mo) includes Astra in ChatGPT Work and Codex, not regular Chat. It draws on a limited Work/Codex allowance, which OpenAI says Astra can use faster than GPT-5.6 Sol.

Is GPT-6 Pro the same as GPT-6 Astra?

GPT-6 Pro is Astra's name in regular Chat on the Pro plan, with its own message limits. Notebookcheck reports 50 a week on Pro at $100/mo and 200 on Pro at $200/mo.

How much does GPT-6 Astra cost?

On the OpenAI API, $10 per million input tokens and $50 per million output, with $20 / $75 for prompts over 272K tokens. On TulexAI, 25 Platform Tokens per 750 tokens of prompt plus reply on Pro ($23/mo), VIP and Elite.

Is GPT-6 Astra better than Claude Fable 5.1?

On Artificial Analysis' Intelligence Index v4.3 they tie at 53, and at 62 on the Coding Agent Index. Astra got there at $3.26 per task versus Fable's $7.63. Fable's edge on TulexAI is tool support.

Does GPT-6 Astra support function calling?

On the OpenAI API, only through the Responses API; Chat Completions does not support it. On TulexAI, Astra runs without tools, so use Claude Fable 5.1 or GPT-5.6 Sol for tool calling.

Can I use GPT-6 Astra without a ChatGPT subscription?

Yes. It's on the OpenAI API, Amazon Web Services, and TulexAI from the Pro plan ($23/mo).

Create your TulexAI account, choose Pro or higher, and run your hardest prompt through GPT-6 Astra, then Claude Fable 5.1, on one subscription.