GPT-6.2 Astra

Released

GPT-6.2 Astra is an AI model released by OpenAI on Tuesday, Nov 17 2026, 12 days after GPT-6 Cyber. It has a 1.05M token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover Blocker Deception (OpenAI), Simulated Workplace Tasks (OpenAI), BenchCAD 2.1 Core, SecureSWE, PatchBench, Terminal-Bench-Science 0.1, and 13 more.

API pricing

Up to 272K input tokens
Input
$8.00
Cached input
$0.40
Cache write
$10.00
Output
$40.00
Over 272K input tokens
Input
$16.00
Cached input
$0.80
Cache write
$20.00
Output
$60.00
  • Reasoning or thinking is supported.
  • OpenAI also publishes a Fast mode for this model at exactly twice the standard rate in every column — $16.00 per million input tokens and $80.00 per million output at short context.
  • At standard rates, 12-hour cache writes cost $20 per million tokens up to 272K input tokens and $40 above that threshold.
  • Ultrafast costs 3× the standard rate through February 28, 2027 and 6× starting March 1, 2027 (Pacific time).
  • Eligible Responses API requests in the background completion beta receive 30% off standard model token rates with a 2-hour window or 40% off with an 8-hour window.
USD per 1M tokensModel mappingVerified November 17, 2026Read from developers.openai.com. Click the check to open the source.

Benchmarks

Coding

BenchCAD 2.1 CoreCAD programming — partial-credit score for modeling hardware designs in the held-out Core evaluationvia OpenAI GPT-6.2 launch
82.5%
#1
SecureSWESecure code generation — OpenAI’s internal evaluation of writing code that is both functional and securevia OpenAI GPT-6.2 launch
65.7%
#1
PatchBenchSecurity patching — security patches that fix vulnerabilities while preserving intended behaviorvia OpenAI GPT-6.2 launch
79.7%
#1
FrontierCode 2.0Agentic coding — whether a model’s code changes to open-source repositories would be merged by their maintainers (version 2.0); separate from FrontierCode v1.1 scoresvia OpenAI GPT-6.2 launch
44.2%
#3 of 8

Terminal & CLI

Terminal-Bench-Science 0.1Agentic scientific computing — The same command-line setup as Terminal-Bench, pointed at scientific work: the AI has to drive research tooling and computational workflows through to a result, rather than administer a machine. Version 0.1 is the first release of the task set, and scores run lower than on the general board. Higher is better.via OpenAI GPT-6.2 launch
77.1%
#1

Agentic & tool use

Agents’ Last Exam v1.1Agentic computer use — averaged rubric score on the 122-task v1.1 suite of long-horizon professional workflowsvia OpenAI GPT-6.2 launch
78.3%
#1
AutomationBench 1.0.6Business workflows — end-to-end business workflow automation on the version 1.0.6 task setvia OpenAI GPT-6.2 launch
52.4%
#1
LifeValEveryday tasks — OpenAI’s rubric score for completing practical everyday tasks tailored to a person’s needsvia OpenAI GPT-6.2 launch
86.1%
#1

Reasoning & science

GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.via OpenAI GPT-6.2 system card
97%
#1
Vals Science Index v1Scientific reasoning — Vals AI’s composite accuracy across three scientific reasoning evaluationsvia OpenAI GPT-6.2 launch
68.4%
#1

Knowledge work

AA Intelligence Index v5.0.1Overall intelligence — Artificial Analysis’s version 5.0.1 composite across nine evaluations of knowledge work, coding, science, and general knowledgevia OpenAI GPT-6.2 launch
62.1
#1
GDPval v2 (OpenAI Internal)Knowledge work — OpenAI’s internal score on multistep professional assignments drawn from consulting, finance, and design projectsvia OpenAI GPT-6.2 launch
50.4%
#1

Finance

AA-AnalystAgent (pass^5)Financial research — share of quantitative questions about real spreadsheets and reports answered correctly on all five independent attemptsvia OpenAI GPT-6.2 launch
62.5%

Healthcare

HealthBench ProfessionalHealth — unadjusted physician-rubric score using the official OpenAI grading procedurevia OpenAI GPT-6.2 system card
66%OpenAI grading, raw
66.3%OpenAI grading, length-adjusted
OpenAI grading, raw
#2 of 5
OpenAI grading, length-adjusted
#1
MentalHealthBenchMental health — overall clinician-rubric score on mental health conversationsvia OpenAI GPT-6.2 system card
53.3%
#1

Robustness

Blocker Deception (OpenAI)Blocker disclosure — share of coding trials where the model omitted or misrepresented an obstacle without the Codex developer promptvia OpenAI GPT-6.2 system card
22.28%without Codex prompt
8.04%with Codex prompt
without Codex prompt
#1
with Codex prompt
#1
Simulated Workplace Tasks (OpenAI)High-severity improper actions — share of scored simulated workplace tasks with a high-severity improper actionvia OpenAI GPT-6.2 system card
0.64%
#1

Community preference

AA-3DCommunity preference (interactive 3D) — blind human preferences between paired interactive 3D experiences built from the same promptvia OpenAI GPT-6.2 launch
1318
#1

Source: OpenAI GPT-6.2 launch, retrieved 17 November 2026 · Source: OpenAI GPT-6.2 system card, retrieved 17 November 2026. Other sources are identified on the linked benchmark pages.

Compare GPT-6.2 Astra with

GPT-6.2 Astra

About

GPT-6.2 Astra was introduced on November 17, 2026 as OpenAI's flagship for complex reasoning and coding, succeeding GPT-6 Astra. The launch post positioned it for hard professional projects: software, computer use, and slides, spreadsheets and documents that follow a supplied template. OpenAI made it available the same day to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.2-astra. Pro 500 subscribers could use Ultrafast. Ultrafast for Pro 200 subscribers started with GPT-6.2 Sol.

The API accepts text and images and returns text. It has a 1,050,000-token context window, with limits of 922,000 input and 128,000 output tokens. Its default context in Codex is 272,000 tokens. The knowledge cutoff is May 18, 2026, and reasoning effort ranges from low through maximum. Both Responses and Chat Completions are available, with tool calling through Responses. Standard short-context rates were $8 per million input tokens, $0.40 cached and $40 output.

OpenAI reported 78.3% on Agents' Last Exam v1.1 at xhigh effort and 77.1% on Terminal-Bench-Science 0.1 at maximum effort. On its internal GDPval v2, Astra scored 6.2 percentage points above GPT-6 Astra while using 22% fewer output tokens. Under OpenAI's Preparedness Framework, Astra met Critical for cybersecurity and High for biological and chemical capabilities. OpenAI described it as the first externally released model to meet High for AI self-improvement. At launch, approved Daybreak users continued to use GPT-6 Astra on Blue and GPT-6 Cyber on Red.

Frequently asked questions

GPT-6.2 Astra was released by OpenAI on Tuesday, Nov 17 2026.

All OpenAI releases

57 tracked

2022

1 release