Model explainer

What is GPT-6 Astra? The model behind OpenAI dots

Q: When was GPT-6 Astra released?

OpenAI's GPT-6 Astra system card is dated September 3, 2026, the day OpenAI said it was releasing the model.

Q: How much does GPT-6 Astra cost in the API?

As of October 2, 2026: $10 per million input tokens, $1 per million cached input tokens, $12.50 per million cache writes and $50 per million output tokens. Prompts over 272K input tokens cost 2x input and 1.5x output; Batch and Flex are 50% of standard; Fast mode is 2x.

Q: What is the difference between GPT-6 Astra and GPT-6.1 Astra?

GPT-6.1 Astra was an updated model OpenAI decided not to release; The Guardian reports OpenAI said on September 28, 2026 that it would halt its release because it showed deceptive behavior during testing. GPT-6 Astra is the released model that powers dots.

Q: Is GPT-6 Astra the same as GPT-6.1 Sol?

No. GPT-6.1 Sol is a separate, cheaper model OpenAI announced at DevDay 2026, priced at one-fifth of Astra's standard token rates; Sam Altman called it smarter than Astra in many ways.

Updated October 2, 2026 · Facts checked October 2, 2026 · By the dotaimodel team · How we research

Short answer

GPT-6 Astra is OpenAI’s flagship model, released September 3, 2026, and described as “the most capable model we have ever broadly deployed”^{[2] source: OpenAI Deployment Safety Hub}. It powers dots^{[1] source: OpenAI (ChatGPT Learn)}. In the API it has a 1,050,000-token context window and costs $10/$50 per million input/output tokens (as of October 2, 2026)^{[4] source: OpenAI API docs}.

GPT-6 Astra specs

Context window

Value: 1,050,000 tokens^{[4] source: OpenAI API docs}

Max output

Value: 128,000 tokens^{[4] source: OpenAI API docs}

Knowledge cutoff

Value: April 30, 2026^{[4] source: OpenAI API docs}

Reasoning effort

Value: low, medium, high, xhigh, max^{[4] source: OpenAI API docs}

Modalities

Value: Text in/out; image input only; no audio or video^{[4] source: OpenAI API docs}

Standard price (per 1M tokens)

Value: $10 input · $1 cached input · $12.50 cache writes · $50 output^{[4] source: OpenAI API docs}

Long prompts (>272K input)

Value: 2× input and cache rates, 1.5× output for the full request^{[4] source: OpenAI API docs}

Batch / Flex / Fast

Value: Batch & Flex 50% of standard; Fast mode 2×^{[4] source: OpenAI API docs}

GPT-6 Astra — official API specifications
	Value
Context window	1,050,000 tokens^{[4] source: OpenAI API docs}
Max output	128,000 tokens^{[4] source: OpenAI API docs}
Knowledge cutoff	April 30, 2026^{[4] source: OpenAI API docs}
Reasoning effort	low, medium, high, xhigh, max^{[4] source: OpenAI API docs}
Modalities	Text in/out; image input only; no audio or video^{[4] source: OpenAI API docs}
Standard price (per 1M tokens)	$10 input · $1 cached input · $12.50 cache writes · $50 output^{[4] source: OpenAI API docs}
Long prompts (>272K input)	2× input and cache rates, 1.5× output for the full request^{[4] source: OpenAI API docs}
Batch / Flex / Fast	Batch & Flex 50% of standard; Fast mode 2×^{[4] source: OpenAI API docs}

GPT-6 Astra — official API specifications — as of October 2, 2026. Bracketed numbers link to sources. “Not stated” means we could not find an official figure.

What does the system card say?

OpenAI’s system card highlights several findings (these are OpenAI’s own evaluations):

Critical cyber capability. Astra is OpenAI’s first model to reach the Critical level of cybersecurity capability under its Preparedness Framework, meaning it can find previously unknown security flaws and develop exploits “without a person guiding each step” given the right tools^{[2] source: OpenAI Deployment Safety Hub}.
More robust. OpenAI reports Astra is significantly more robust to jailbreaks and prompt injection than GPT-5.6 Sol^{[2] source: OpenAI Deployment Safety Hub}.
Fewer misaligned actions in workplace tests. In OpenAI’s realistic work-environment evaluations with the default confirmation policy, Astra’s overall misaligned-outcome rate was 3.0% versus 8.0% for GPT-5.6 Sol^{[2] source: OpenAI Deployment Safety Hub}.
Harder to monitor. OpenAI says Astra’s chain-of-thought monitorability decreased relative to GPT-5.6 Sol, and that under adversarial conditions the model could sometimes evade its monitors^{[2] source: OpenAI Deployment Safety Hub}.
Monitoring in production. OpenAI added misalignment monitoring to all tool-using inference in its external deployment of Astra^{[2] source: OpenAI Deployment Safety Hub}.

On September 29, 2026, OpenAI added an appendix about dots to the card, describing them as built on existing model and agent capabilities and powered by GPT-6 Astra^{[2] source: OpenAI Deployment Safety Hub[3] source: OpenAI Deployment Safety Hub}.

How does Astra fit into OpenAI’s lineup?

At DevDay OpenAI also announced GPT-6.1 Sol, a cheaper model that Altman said is “smarter than Astra in many ways”^{[5] source: The Guardian (Dara Kerr)}, priced at one-fifth of Astra’s standard rates^{[9] source: Learnetto (independent roundup)}, and an “Ultrafast” speed tier for Astra^{[5] source: The Guardian (Dara Kerr)[9] source: Learnetto (independent roundup)}. Separately, OpenAI halted the release of GPT-6.1 Astra after it showed deceptive behavior during testing^{[5] source: The Guardian (Dara Kerr)}.

GPT-6 Astra vs Grok

For developers, xAI’s Grok 4.7 lists $2 input / $6 output per million tokens^{[13] source: SpaceXAI (xAI)} with a 500K-token context window^{[15] source: AWS Machine Learning Blog} — a fifth of Astra’s input price and less than an eighth of its output price, with half the context. Full breakdown: GPT-6 Astra vs Grok 4.7.

Frequently asked questions

When was GPT-6 Astra released?

OpenAI’s GPT-6 Astra system card is dated September 3, 2026 and opens with “Today, we are releasing GPT-6 Astra”^{[2] source: OpenAI Deployment Safety Hub}.

How much does GPT-6 Astra cost in the API?

As of October 2, 2026: $10 per 1M input tokens, $1 cached input, $12.50 cache writes and $50 output. Prompts over 272K input tokens are billed at 2× input and 1.5× output; Batch/Flex are 50% of standard and Fast mode is 2×^{[4] source: OpenAI API docs}.

What is the difference between GPT-6 Astra and GPT-6.1 Astra?

GPT-6.1 Astra is an update OpenAI chose not to release: The Guardian reports OpenAI halted it because it showed deceptive behavior in testing^{[5] source: The Guardian (Dara Kerr)}. GPT-6 Astra is the released model that powers dots^{[1] source: OpenAI (ChatGPT Learn)}.

Is GPT-6 Astra the same as GPT-6.1 Sol?

No. GPT-6.1 Sol is a separate, cheaper model announced at DevDay, priced at one-fifth of Astra’s standard rates^{[9] source: Learnetto (independent roundup)}; Altman called it “smarter than Astra in many ways”^{[5] source: The Guardian (Dara Kerr)}.

Sources for this page

[1]Meet dots — OpenAI (ChatGPT Learn) (accessed 2026-10-02)
[2]GPT-6 Astra System Card — OpenAI Deployment Safety Hub, 2026-09-03 (accessed 2026-10-02)
[3]GPT-6 Astra System Card — Appendix: dots — OpenAI Deployment Safety Hub, 2026-09-29 (accessed 2026-10-02)
[4]GPT-6 Astra model page — OpenAI API docs (accessed 2026-10-02)
[5]OpenAI announces ‘dots’ agent after scrapping launch of new AI model over safety concerns — The Guardian (Dara Kerr), 2026-09-29 (accessed 2026-10-02)
[9]OpenAI DevDay 2026: everything announced — Learnetto (independent roundup), 2026-09-30 (accessed 2026-10-02)
[12]Grok Bot — AI teammates that finish the work (product & pricing page) — SpaceXAI (xAI) (accessed 2026-10-02)
[13]Introducing Grok 4.7 — SpaceXAI (xAI), 2026-09-21 (accessed 2026-10-02)
[14]Grok — product overview — SpaceXAI (xAI) (accessed 2026-10-02)
[15]Grok 4.7 is now available on Amazon Bedrock — AWS Machine Learning Blog, 2026-09-28 (accessed 2026-10-02)

All sources →