OpenAI logo

GPT-6.2 Astra

Multimodal
OpenAI

GPT-6.2 Astra is OpenAI's frontier GPT-6.2 model for the hardest end-to-end professional work, including complex coding, computer use, scientific work, and creating slides, spreadsheets, and documents. The API model id is gpt-6.2-astra. It accepts text and image input with text output, supports low, medium, high, xhigh, and max reasoning effort (but not none), and provides a 1,050,000-token context window (922,000 max input) with up to 128,000 output tokens. Responses API tools include web and file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search; it also supports streaming, function calling, structured outputs, and multi-agent delegation (beta). Standard per-million-token pricing is $8 input, $0.40 cached input, $10 cache writes (30-minute retention), and $40 output, 20% below GPT-6 Astra on input and output, with cached input 60% below GPT-6 Astra ($0.40 vs $1.00). Prompts above 272K input tokens cost 2x input/cache and 1.5x output across the full request. Batch costs 50% of Standard; Fast costs 2x; Ultrafast costs 3 times the Standard rate through February 28, 2027, and 6 times the Standard rate starting March 1, 2027 (Pacific time). Launch benchmarks are OpenAI-reported; effort variants are recorded with each score.

Key Specifications

Parameters
-
Context
1.1M
Release Date
November 17, 2026
Average Score
72.3%

Timeline

Key dates in the model's history
Announcement
November 17, 2026
Last Update
November 18, 2026
Today
November 20, 2026

Technical Specifications

Parameters
-
Training Tokens
-
Knowledge Cutoff
May 18, 2026
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$8.00
Output (per 1M tokens)
$40.00
Max Input Tokens
1.1M
Max Output Tokens
128.0K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis
GPQA
GPQA Diamond (198 questions, no tools), accuracy 97%. GPT-6.2 Astra and Sol system card, Table 2 (High biological capability evaluations). • Self-reported
97.0%

Other Tests

Specialized benchmarks
Agents' Last Exam
Agents' Last Exam v1.1. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: xhigh effort (0.7833). Other efforts: low 0.6577, medium 0.7206, high 0.7596, max 0.7761. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
78.3%
AutomationBench v1.0.6
AutomationBench 1.0.6. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.524). Other efforts: low 0.371, medium 0.438, high 0.479, xhigh 0.472. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
52.4%
ExploitBench
ExploitBench V8, mean capability credit (five attempts per vulnerability) 100% at every reasoning setting. GPT-6.2 Astra and Sol system card, Table 4 (Critical cybersecurity capability evaluations). • Self-reported
100.0%
FrontierCode 2.0
Cognition FrontierCode 2.0. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.442). Other efforts: low 0.381, medium 0.422, high 0.436, xhigh 0.441. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
44.2%
GDPval v2
GDPval v2 (OpenAI internal). OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.504). Other efforts: low 0.289, medium 0.393, high 0.45, xhigh 0.478. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
50.4%
HealthBench Professional
HealthBench Professional, length-adjusted score (unadjusted 0.660, mean visible response length 1,884 characters). GPT-6.2 Astra and Sol system card, Table 17. • Self-reported
66.3%
Internal Research Debugging Evaluation
Internal Research Debugging Evaluation, mean rubric score 81.85%. GPT-6.2 Astra and Sol system card, Table 5 (Research engineering and model-training results). • Self-reported
81.8%
KernelGen 1P
KernelGen 1P, mean rubric score 76.92%. GPT-6.2 Astra and Sol system card, Table 5 (Research engineering and model-training results). • Self-reported
76.9%
LifeVal
LifeVal. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.861). Other efforts: low 0.463, medium 0.789, high 0.831, xhigh 0.837. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
86.1%
NanoGPT
NanoGPT, mean share of baseline training time saved 42.87%. GPT-6.2 Astra and Sol system card, Table 5. The system card notes NanoGPT uses slightly different evaluation versions for the two Astra models. • Self-reported
42.9%
PatchBench
PatchBench. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: xhigh effort (0.797). Other efforts: low 0.711, medium 0.756, high 0.788, max 0.797. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
79.7%
PostTrainBench Lite
PostTrainBench Lite, mean share of performance gap closed 74.79%. GPT-6.2 Astra and Sol system card, Table 5 (Research engineering and model-training results). • Self-reported
74.8%
SEC-bench Pro
SEC-Bench Pro, vulnerability reproduction (pass@1) 86.75% at maximum reasoning effort. GPT-6.2 Astra and Sol system card, Table 4 (Critical cybersecurity capability evaluations). • Self-reported
86.8%
SecureSWE
SecureSWE (OpenAI internal). OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.657). Other efforts: low 0.545, medium 0.577, high 0.617, xhigh 0.638. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
65.7%
Terminal-Bench-Science 0.1
Terminal-Bench-Science 0.1. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.771). Other efforts: low 0.563, medium 0.723, high 0.749, xhigh 0.74. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
77.1%
Vals Science Index
Vals Science Index v1. OpenAI GPT-6.2 Astra launch interactive chart. Highest published score across the reported effort sweep: max effort (0.684). Other efforts: low 0.595, medium 0.676, high 0.681, xhigh 0.648. Evaluations used OpenAI research environments or API and may differ from production ChatGPT. • Self-reported
68.4%

License & Metadata

License
proprietary
Announcement Date
November 17, 2026
Last Updated
November 18, 2026

Compare GPT-6.2 Astra

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.