Skip to content
← Models
ClosedOpenAI/10 trustVerified

GPT-6 Astra

OpenAI's top-tier frontier reasoning model for computer use, agentic coding, and long-horizon professional work.

Visit official page →

About

GPT-6 Astra is OpenAI's current flagship model, succeeding GPT-5.6 Sol, with strengths in computer use, terminal-based science work, and long-context retrieval. OpenAI classifies it at the "Critical" cybersecurity capability level under its Preparedness Framework, shipping with strengthened safeguards.

It sits above GPT-6.1 Sol / GPT-6 Sol / GPT-6 Luna in the current seven-model lineup. (Note: GPT-6.1 Astra was canceled by OpenAI in October 2026.)

Specs

Context window

1,050,000 tokens (922,000 max input; 128,000 max output)

Pricing

$10 input / $50 output per 1M tokens; $1 per 1M cached input; $12.50 per 1M cache writes. Requests above 272K input tokens billed at 2x input/cache and 1.5x output. Batch and Flex at 50% of standard.

Free quota

None on the API. ChatGPT Free/Go accounts get the smaller GPT-5.6 Luna as default with uncapped text messaging; Astra is available via ChatGPT Plus and Pro/Business/Enterprise (rolling out Oct 2026).

API access

OpenAI API (Responses and Chat Completions), AWS Bedrock, Microsoft Azure

Weights

Closed — API only

Released

September 3, 2026

Benchmarks

OpenAI-reported launch figures:

  • ARC-AGI-3: 99.9%
  • OSWorld 2.0: 72.6%
  • Terminal-Bench Science 0.1: 64.6%
  • SRE-Bench: 88% pass@1

Independent: Artificial Analysis Intelligence Index 53 (Oct 2026).

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

chatcodevisionreasoningautomationdata

Trust score — 8/10

/10 trustVerified

OpenAI is an established provider with strong docs, clear published pricing (even the long-context price lane is documented), and a benchmark table that includes its own losses — unusually transparent. Caveats: all launch scores are provider-reported at max reasoning effort rather than same-cost comparisons, and the 'Critical' cyber rating means safety stops can interrupt agentic tasks.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
5.1/10
Manual review · sanity read + red-flag check
9.0/10
  • · provider-tier estimate — thinner evidence (no public adoption metric)
  • · released 4 days ago
  • · manual review: max-effort benchmarks; safety eval stops at Opus 4.6

Sources