📊 See AI model rankings

📱 Get the AI Rank app

GPT-6 Astra is OpenAI’s GPT-6 flagship. The company calls it its most intelligent and aligned model to date, with state-of-the-art claims across computer use, browsing, software engineering, cybersecurity, science, and professional work. API id: gpt-6-astra. Standard API pricing on the launch page: $10 per million input tokens and $50 per million output tokens (separate rates for cache reads/writes; Fast mode up to 2× speed at 2× Standard price). Cost-focused sibling: GPT-6.1 Sol.

Availability

OpenAI’s Astra page says the model rolled out first to a limited set of organizations, then to ChatGPT Plus, Pro, Business, and Enterprise users, plus the OpenAI API, Microsoft Azure, and AWS Bedrock. Enterprise admins enable it (off by default at launch). Pro/Business/Enterprise also get GPT-6 Astra Pro. Zero Data Retention is mentioned for eligible API customers.

Cybersecurity: Critical Threshold

OpenAI states Astra meets the Critical cybersecurity threshold under its Preparedness Framework. Launch-page cyber figures (without production safeguards in the cited setups) include ExploitBench 100%, ExploitGym 42.4%, and strong SRE-Bench scores versus GPT-5.6 Sol. The shipping product refuses advanced offensive requests such as creating proof-of-concept exploits; OpenAI points to Daybreak for broader defensive access later. Read the Astra safety overview / system card for the full control stack—do not treat benchmark rows as permission to misuse the model.

Selected Official Eval Highlights

From the Astra announcement tables (max over effort; research/API harness caveats apply):

  • ARC-AGI-3: 99.9%
  • ExploitBench: 100%
  • OSWorld 2.0 offline partial: 72.6%
  • Terminal-Bench 4.0: 57.9%
  • DeepSWE v1.1: 74.1%
  • BenchCAD (with tools): 95.9%

Other science/math claims on the page include FrontierMath Tier 4 saturation language and prime-gap research notes—open those linked proofs if you need the math detail.

Alignment Notes (Official)

OpenAI highlights lower rates of going beyond authorized scope versus GPT-5.6 Sol on an internal “impossible task” eval (0% vs 48% in the setup they describe), stronger respect for Auto-Review denials, and improved capability-hallucination metrics. Production still adds monitoring and Auto-Review style pauses that can interrupt legitimate defensive work.

FAQ

What is the GPT-6 Astra API name?

gpt-6-astra

How much does the API cost?

$10 / $50 per million input/output tokens at Standard rates on the launch page.

How is Sol different?

GPT-6.1 Sol is priced at $2 / $10 with OpenAI’s “near-Astra” framing and is not yet in Chat.

Sources