name = "GPT-6 Astra"
description = "GPT-6 Astra is OpenAI's most capable model for complex reasoning, coding, computer use, research, and document creation."
family = "gpt-astra"
release_date = "2026-09-04"
last_updated = "2026-09-04"
attachment = true
reasoning = true
temperature = false
tool_call = true
structured_output = true
knowledge = "2026-04-30"
open_weights = false

[limit]
context = 1_050_000
input = 922_000
output = 128_000

[modalities]
input = ["text", "image", "pdf"]
output = ["text"]

[[benchmarks]]
name = "Agents' Last Exam"
score = 59.3
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "OSWorld"
score = 72.6
metric = "partial score"
dataset = "V2-Offline (v2026.08.08)"
version = "2.0"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "ScreenSpot-Pro"
score = 92.7
metric = "accuracy"
variant = "no tools"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "AutomationBench"
score = 41.4
metric = "success rate"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "BenchCAD"
score = 95.9
metric = "geometric overlap"
variant = "with tools"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "BrowseComp"
score = 91.5
metric = "accuracy"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "Terminal-Bench"
score = 57.9
metric = "success rate"
version = "4.0"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "DeepSWE"
score = 74.1
metric = "resolve rate"
version = "1.1"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "FrontierCode"
score = 64.5
metric = "score"
dataset = "Extended"
version = "1.1"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "Terminal-Bench Science"
score = 64.6
metric = "success rate"
version = "0.1"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "FrontierMath"
score = 97.6
metric = "accuracy"
dataset = "Tier 4"
version = "v2"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "GPQA Diamond"
score = 96.0
metric = "accuracy"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 57.2
metric = "accuracy"
variant = "with tools"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "ExploitBench"
score = 100.0
metric = "success rate"
variant = "without production safeguards"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "SRE-Bench"
score = 88.0
metric = "success rate"
variant = "single attempt"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "ARC-AGI-3"
score = 99.9
metric = "RHAE"
harness = "Responses API"
source = "https://openai.com/index/gpt-6-astra/"
date = "2026-09-03"

[[benchmarks]]
name = "Artificial Analysis Intelligence Index"
score = 61
metric = "index score"
variant = "max"
version = "4.1.1"
source = "https://artificialanalysis.ai/articles/benchmarking-gpt-6-astra"
date = "2026-09-03"

[[benchmarks]]
name = "Artificial Analysis Coding Agent Index"
score = 67
metric = "index score"
harness = "Codex"
variant = "max"
version = "1.4"
source = "https://artificialanalysis.ai/articles/benchmarking-gpt-6-astra"
date = "2026-09-03"