GPT-6 Astra

OpenAICurrent

Input price: $10.00; Per 1 million tokens (USD)
Output price: $50.00; Per 1 million tokens (USD)
Effort levels: 5total; Default not stated
Benchmark results: 55total; Benchmark types: 26

Model information

API model ID: gpt-6-astra
release date: 2026-09-03
effort: Low / Med / High / Xhigh / Max
Effort source: Based on reasoning effort on the OpenAI model page (developers.openai.com/api/docs/models/gpt-6-astra).
Claude Code: unavailable
Codex: available
Availability notes: Codex models page: 'codex -m gpt-6-astra'. Codex UI labels are Light (=low)/Medium/High/Extra High (+Max/Ultra). 'Some paid plans omit Astra Extra High.' none is not supported. The API default effort is not stated on the model page. Claude Code's official model list (code.claude.com/docs/en/model-config alias table) contains only Anthropic models, so claudeCode is false (connections via a gateway are not counted as official options).
Availability source: https://learn.chatgpt.com/docs/models

Price (USD per 1 million tokens)

Prices do not vary by effort; token consumption does.

input	$10.00
output	$50.00
cache read	$1.00
cache write	$12.50

Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request. Batch/Flex 50%, Fast 2x. Table columns: short input/output/cached/30m write/12h write, followed by the same columns for >272K (input $20, output $75, cached $2). cacheWrite is 30m write.

| gpt-6-astra | $10.00 | $50.00 | $1.00 | $12.50 | $25.00 | $20.00 | $75.00 | $2.00 | $25.00 | $50.00 |

Source: https://developers.openai.com/api/docs/pricing

Benchmark results

Benchmark	effort	Score	Source cost (metric)	measured/reported by	source	Conditions and extraction method
Terminal-Bench 4.0Low	Low	49.7%	$4.95Cost per attempt	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Low" was mapped to low in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"transcribed from text Source excerpt Terminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Low: 49.7% $4.95"
Terminal-Bench 4.0Med	Med	53.9%	$6.15Cost per attempt	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"transcribed from text Source excerpt Terminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Med: 53.9% $6.15"
Terminal-Bench 4.0High	High	57.9%	$7.21Cost per attempt	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "High" was mapped to high in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"transcribed from text Source excerpt Terminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · High: 57.9% $7.21"
Terminal-Bench 4.0Xhigh	Xhigh	57.6%	$7.48Cost per attempt	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"transcribed from text Source excerpt Terminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Xhigh: 57.6% $7.48"
Terminal-Bench 4.0Max	Max	56.7%	$10.35Cost per attempt	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Max" was mapped to max in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"transcribed from text Source excerpt Terminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Max: 56.7% $10.35"
Terminal-Bench-Science 0.1not stated	not stated	64.6%	—	OpenAI	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. table note: Unless otherwise noted, all Claude Opus 5.5 results use adaptive thinking at max effort. production safeguards enabled. Values for other vendor models (as listed by Anthropic; null where effort was not stated). SE ±3.5–5pt. The GPT-6 Astra figure is a reported value from OpenAI.transcribed from table Source excerpt 64.6%Location Table, Terminal-Bench-Science 0.1 row: 64.6
Terminal-Bench-Science 0.1not stated	not stated	68.1%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 68.1%Location Table, Terminal-Bench-Science 0.1 row: 68.1
FrontierCode v1.1(main set)Low	Low	45.3%	$1.59Cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode v1.1, main set / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Low" was mapped to low in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode v1.1, main set — Accuracy vs Cost: "GPT-6 Astra · Low: 45.3% $1.59"
FrontierCode v1.1(main set)Med	Med	48.8%	$2.28Cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode v1.1, main set / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode v1.1, main set — Accuracy vs Cost: "GPT-6 Astra · Med: 48.8% $2.28"
FrontierCode v1.1(main set)High	High	50.9%	$2.85Cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode v1.1, main set / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "High" was mapped to high in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode v1.1, main set — Accuracy vs Cost: "GPT-6 Astra · High: 50.9% $2.85"
FrontierCode v1.1(main set)Xhigh	Xhigh	50.6%	$3.10Cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode v1.1, main set / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode v1.1, main set — Accuracy vs Cost: "GPT-6 Astra · Xhigh: 50.6% $3.10"
FrontierCode v1.1(main set)Max	Max	53.3%	$4.36Cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode v1.1, main set / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Max" was mapped to max in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode v1.1, main set — Accuracy vs Cost: "GPT-6 Astra · Max: 53.3% $4.36"
FrontierCode 2.0Low	Low	31.1%	$2.44Cost per task	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode 2.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Low" was mapped to low in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode 2.0 — Accuracy vs. cost: "GPT-6 Astra · Low: 31.1% $2.44"
FrontierCode 2.0Med	Med	36.8%	$3.49Cost per task	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode 2.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode 2.0 — Accuracy vs. cost: "GPT-6 Astra · Med: 36.8% $3.49"
FrontierCode 2.0High	High	38.3%	$4.28Cost per task	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode 2.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "High" was mapped to high in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode 2.0 — Accuracy vs. cost: "GPT-6 Astra · High: 38.3% $4.28"
FrontierCode 2.0Xhigh	Xhigh	41.4%	$4.54Cost per task	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode 2.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode 2.0 — Accuracy vs. cost: "GPT-6 Astra · Xhigh: 41.4% $4.54"
FrontierCode 2.0Max	Max	42.8%	$6.68Cost per task	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "FrontierCode 2.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Max" was mapped to max in the OpenAI API reasoning efforttranscribed from text Source excerpt FrontierCode 2.0 — Accuracy vs. cost: "GPT-6 Astra · Max: 42.8% $6.68"
FrontierSWE 2not stated	not stated	65.5%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 65.5%Location Table, FrontierSWE v2 row: 65.5
GDPval-AA v2.1Low	Low	1,366	$0.85Estimated cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "GDPval-AA v2.1 / Elo vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Low" was mapped to low in the OpenAI API reasoning efforttranscribed from text Source excerpt GDPval-AA v2.1 — Elo vs Cost: "GPT-6 Astra · Low: 1366 $0.85"
GDPval-AA v2.1Med	Med	1,468	$1.82Estimated cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "GDPval-AA v2.1 / Elo vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning efforttranscribed from text Source excerpt GDPval-AA v2.1 — Elo vs Cost: "GPT-6 Astra · Med: 1468 $1.82"
GDPval-AA v2.1High	High	1,485	$2.43Estimated cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "GDPval-AA v2.1 / Elo vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "High" was mapped to high in the OpenAI API reasoning efforttranscribed from text Source excerpt GDPval-AA v2.1 — Elo vs Cost: "GPT-6 Astra · High: 1485 $2.43"
GDPval-AA v2.1Xhigh	Xhigh	1,516	$3.04Estimated cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "GDPval-AA v2.1 / Elo vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning efforttranscribed from text Source excerpt GDPval-AA v2.1 — Elo vs Cost: "GPT-6 Astra · Xhigh: 1516 $3.04"
GDPval-AA v2.1Max	Max	1,542	$4.53Estimated cost per task	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "GDPval-AA v2.1 / Elo vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Max" was mapped to max in the OpenAI API reasoning efforttranscribed from text Source excerpt GDPval-AA v2.1 — Elo vs Cost: "GPT-6 Astra · Max: 1542 $4.53"
AA-Briefcase v1.1(Elo)not stated	not stated	1,569	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 1569Location Table, AA-Briefcase v1.1 (Elo) row: 1569
AA-Briefcase v1.1(rubric)not stated	not stated	52.0%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 52.0%Location Table, AA-Briefcase v1.1 (rubric) row: 52.0
AutomationBenchnot stated	not stated	41.4%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 41.4%Location Table, AutomationBench row: 41.4
AutomationBenchLow	Low	30.3%	$1.08Cost per task	Zapier	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "AutomationBench / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; measured by: Zapier (Opus 5.5 is from a Zapier early access evaluation; the others are from the Zapier public leaderboard); the figure's effort label "Low" was mapped to low in the OpenAI API reasoning efforttranscribed from text Source excerpt AutomationBench — Accuracy vs Cost: "GPT-6 Astra · Low: 30.3% $1.08"
AutomationBenchMed	Med	34.1%	$1.28Cost per task	Zapier	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "AutomationBench / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; measured by: Zapier (Opus 5.5 is from a Zapier early access evaluation; the others are from the Zapier public leaderboard); the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning efforttranscribed from text Source excerpt AutomationBench — Accuracy vs Cost: "GPT-6 Astra · Med: 34.1% $1.28"
AutomationBenchHigh	High	37.1%	$1.45Cost per task	Zapier	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "AutomationBench / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; measured by: Zapier (Opus 5.5 is from a Zapier early access evaluation; the others are from the Zapier public leaderboard); the figure's effort label "High" was mapped to high in the OpenAI API reasoning efforttranscribed from text Source excerpt AutomationBench — Accuracy vs Cost: "GPT-6 Astra · High: 37.1% $1.45"
AutomationBenchXhigh	Xhigh	39.0%	$1.53Cost per task	Zapier	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "AutomationBench / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; measured by: Zapier (Opus 5.5 is from a Zapier early access evaluation; the others are from the Zapier public leaderboard); the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning efforttranscribed from text Source excerpt AutomationBench — Accuracy vs Cost: "GPT-6 Astra · Xhigh: 39.0% $1.53"
AutomationBenchMax	Max	41.4%	$1.77Cost per task	Zapier	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. A point in the figure "AutomationBench / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; measured by: Zapier (Opus 5.5 is from a Zapier early access evaluation; the others are from the Zapier public leaderboard); the figure's effort label "Max" was mapped to max in the OpenAI API reasoning efforttranscribed from text Source excerpt AutomationBench — Accuracy vs Cost: "GPT-6 Astra · Max: 41.4% $1.77"
Humanity's Last Exam (with tools)not stated	not stated	57.2%	—	not stated	Introducing Claude Opus 5.5official · Retrieved 2026-11-12	Vendor-reported value. table note: Unless otherwise noted, all Claude Opus 5.5 results use adaptive thinking at max effort. production safeguards enabled. Values for other vendor models (as listed by Anthropic; null where effort was not stated).transcribed from table Source excerpt 57.2%Location Table, Humanity's Last Exam (with tools) row: 57.2
Humanity's Last Exam (with tools)not stated	not stated	57.2%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 57.2%Location Table, Humanity's Last Exam (with tools) row: 57.2
OSWorld 2.1 (offline set, partial)not stated	not stated	79.5%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 79.5%Location Table, OSWorld 2.1 (offline set, partial) row: 79.5
Chartography (without tools)not stated	not stated	71.0%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 71.0%Location Table, Chartography (no tools) row: 71.0
RoboDojo Simnot stated	not stated	29.8%	—	Anthropic	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic). ‡ highest-available effort, run internally.transcribed from table Source excerpt 29.8%Location Table, RoboDojo Sim row: 29.8
Remote Labor Index(RLI)not stated	not stated	20.8%	—	not stated	Introducing Claude Fable 5.2 and Claude Mythos 5.2official · Retrieved 2026-11-12	Vendor-reported value. table note: Scores for Fable 5.2, Fable 5.1, and Opus 5.5 show results with safeguards on unless otherwise noted. All scores show each model’s highest-available effort unless otherwise noted. Values for other vendor models (as listed by Anthropic).transcribed from table Source excerpt 20.8%Location Table, Remote Labor Index (RLI) row: 20.8
HealthBench Professional (length-adjusted score)not stated	not stated	64.7	—	OpenAI	GPT-6 Astra System Card (Table 6. HealthBench Evaluations)official · Retrieved 2026-11-12	Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoretranscribed from table Source excerpt 64.7Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row
HealthBench Hard (length-adjusted score)not stated	not stated	36.6	—	OpenAI	GPT-6 Astra System Card (Table 6. HealthBench Evaluations)official · Retrieved 2026-11-12	Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoretranscribed from table Source excerpt 36.6Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Hard row
GPQA DiamondMax	Max	95.8%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, gpqa_diamond.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt gpqa_diamond.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.9577020202020202, stderr=0.013696812097020311, Started at=2026-08-30T14:57:43.000Z
FrontierMath Tiers 1-3(v2)Max	Max	93.7%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tiers_1_3_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tiers_1_3_v2.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.9368421052631579, stderr=0.014434038295343787, Started at=2026-08-30T14:57:43.000Z
FrontierMath Tier 4(v2)not stated	not stated	82.9%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (none)". effort is indicated by the suffix _none in Epoch's Model version. Data retrieved 2026-11-12. No effort because the source effort label "none" is not in the official effort list.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_none, Best score (across scorers)=0.8292682926829268, stderr=0.05949419959829497, Started at=2026-08-30T02:22:12.000Z
FrontierMath Tier 4(v2)Low	Low	87.8%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (low)". effort is indicated by the suffix _low in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_low, Best score (across scorers)=0.8780487804878049, stderr=0.051739520574625435, Started at=2026-08-30T02:22:12.000Z
FrontierMath Tier 4(v2)Med	Med	97.6%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (medium)". effort is indicated by the suffix _medium in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_medium, Best score (across scorers)=0.975609756097561, stderr=0.024390243902439022, Started at=2026-08-30T02:22:12.000Z
FrontierMath Tier 4(v2)High	High	97.6%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (high)". effort is indicated by the suffix _high in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_high, Best score (across scorers)=0.976, stderr=0.024, Started at=2026-08-30T02:22:12.000Z
FrontierMath Tier 4(v2)Xhigh	Xhigh	97.6%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (xhigh)". effort is indicated by the suffix _xhigh in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_xhigh, Best score (across scorers)=0.976, stderr=0.024, Started at=2026-08-30T02:22:12.000Z
FrontierMath Tier 4(v2)Max	Max	97.6%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, frontiermath_tier_4_v2.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt frontiermath_tier_4_v2.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.976, stderr=0.024, Started at=2026-08-30T02:22:12.000Z
OTIS Mock AIME 2024-2025Max	Max	100.0%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, otis_mock_aime_2024_2025.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt otis_mock_aime_2024_2025.csv: Model version=gpt-6-astra_max, Best score (across scorers)=1.0, stderr=0.0, Started at=2026-08-30T14:57:43.000Z
SimpleQA VerifiedMax	Max	75.6%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, simpleqa_verified.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt simpleqa_verified.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.756, stderr=0.013588548437881386, Started at=2026-08-30T14:57:43.000Z
Chess Puzzles(Epoch AI)Max	Max	74.0%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, chess_puzzles.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt chess_puzzles.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.74, stderr=0.0440844002276808, Started at=2026-10-13T13:42:38.000Z
MirrorCodeHigh	High	46.7%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, mirrorcode.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (high)". effort is indicated by the suffix _high in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt mirrorcode.csv: Model version=gpt-6-astra_high, Best score (across scorers)=0.467, stderr=0.1186, Started at=2026-08-30T16:51:09.000Z
MirrorCodeMax	Max	60.0%	—	Epoch AI	Epoch AI, Capabilities & benchmarking (benchmark_data.zip, mirrorcode.csv) CC BY 4.0third-party · Retrieved 2026-11-12	Run by Epoch AI (third-party measurement). Source label "GPT-6 Astra (max)". effort is indicated by the suffix _max in Epoch's Model version. Data retrieved 2026-11-12.transcribed from table Source excerpt mirrorcode.csv: Model version=gpt-6-astra_max, Best score (across scorers)=0.6, stderr=0.11818736805705578, Started at=2026-10-29T18:04:06.000Z
Epoch Capabilities Index(ECI)not stated	not stated	169	—	Epoch AI	Epoch AI, epoch_capabilities_index/eci_scores.csv (CC BY 4.0)third-party · Retrieved 2026-11-12	A per-model metric that combines results across multiple benchmarks and multiple efforts (not per effort). Calculated by Epoch AI.transcribed from table Source excerpt eci_scores.csv: Model=GPT-6 Astra, eci=169.0, eci_ci_low=166.08, eci_ci_high=172.26
LMArena Text (style control, overall)Max	Max	1,474	—	LMArena	lmarena-ai/leaderboard-dataset text_style_control/latest-00000-of-00001.parquet (CC BY 4.0)third-party · Retrieved 2026-11-12	Human head-to-head votes (Bradley-Terry). Source label "GPT-6 Astra (Max)" (effort is in parentheses in the label). Release date 2026-11-09.transcribed from table Source excerpt model_name=GPT-6 Astra (Max), rating=1474.09, rating_lower=1468.5, rating_upper=1479.7, vote_count=18315, rank=41, category=overall, leaderboard_publish_date=2026-11-09
LMArena Text Coding (without style control)Max	Max	1,494	—	LMArena	lmarena-ai/leaderboard-dataset text/latest-00000-of-00001.parquet (CC BY 4.0)third-party · Retrieved 2026-11-12	Human head-to-head votes (Bradley-Terry). Source label "GPT-6 Astra (Max)" (effort is in parentheses in the label). Release date 2026-11-09.transcribed from table Source excerpt model_name=GPT-6 Astra (Max), rating=1494.39, rating_lower=1484.6, rating_upper=1504.2, vote_count=4154, rank=33, category=coding, leaderboard_publish_date=2026-11-09

GPT-6 Astra

Model information

Price (USD per 1 million tokens)

Benchmark results

Related rankings

References