Inference
EN
OpenAI

GPT-6 Astra

GPT-6 Astra is a reasoning model from OpenAI in the GPT-6 family. 76 benchmarks count toward its index, in 8 categories.

  • Closed weights
  • Reasoning
  • Input: file · image · text
  • Released 2026-09
RankedEnough results for an index.
IndexAll categories, out of 100.78.0±1.9Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.34/sMedian 67/s
First tokenSeconds to the first output token.3.8 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$10.00batch $5.00Median $0.60
Output / 1MUS dollars per 1M output tokens.$50.00batch $25.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1440±6 #77Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-6-astra
Max outputMaximum output tokens per request.
128K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$1.00
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
5 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Amazon BedrockCloud$10.00$50.001.05M
DigitalOceanCloud$10.00$50.001.05M
Microsoft AzureCloud$10.00$50.001.05M
OpenAIMaker$10.00$50.001.05M
OpenRouterRouter$10.00$50.001.05M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.33 counted74.69 / 218
CodingCode writing and repair.14 counted76.97 / 209
ReasoningLogic problems and puzzles.11 counted83.83 / 227
MultimodalTasks with images and text.4 counted71.29 / 84
KnowledgeFacts and expert knowledge.7 counted78.54 / 268
MultilingualTasks in many languages.1 counted73.6—
InstructionTasks with strict prompt rules.1 counted79.5—
MathMath problems.5 counted78.03 / 168

Results

76 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
OTIS Mock AIME 2024-2025Math100.0%1 / 200
Artificial Analysis GPQA DiamondKnowledge96.1%1 / 199
EuroEval SwedishMultilingual77.6% ±0.51 / 195
EuroEval GermanMultilingual68.0% ±0.51 / 191
EuroEval SpanishMultilingual70.0% ±0.41 / 176
EuroEval DutchMultilingual74.7% ±0.31 / 175
Graduate-Level Google-Proof Q&AKnowledge96.0%1 / 141
ARC-AGI-2 (semi-private)Reasoning95.0%1 / 131
EuroEval PolishMultilingual67.4% ±0.42 / 209
EuroEval PortugueseMultilingual74.7% ±0.42 / 182
EuroEval FrenchMultilingual76.3% ±0.42 / 178
EuroEval ItalianMultilingual77.8% ±0.32 / 174
GPQA Diamond · BenchLMKnowledge96.0%1 / 84
GPQA diamond · Epoch AIKnowledge95.8% ±1.43 / 217
Artificial Analysis MMMU-ProMultimodal86.9%2 / 113
Critical Physics TasksReasoningAbove fitFar above this model's index.31.7%4 / 214
MazeBench: visual spatial reasoning in a 3D open worldAgenticAbove fitFar above this model's index.23.0%1 / 48
IOICoding100.0%1 / 45
FrontierMath-Tiers-1-3-v2-PrivateMath93.7% ±1.42 / 90
LiveBench ReasoningReasoning92.7%2 / 80
Mystery Game PuzzlesReasoningAbove fitFar above this model's index.84.0% ±3.72 / 80
LiveBench Data AnalysisReasoning83.0%2 / 80
Terminal-Bench 2.1 · Vals AIAgentic87.3% ±0.42 / 75
Chess PuzzlesReasoning74.0% ±4.43 / 90
Artificial Analysis Humanity's Last ExamKnowledge54.7%8 / 222
Artificial Analysis Omniscience AccuracyKnowledge61.6%6 / 164
Code MigrationCoding67.7% ±4.23 / 77
ARC-AGI-1 (semi-private)Reasoning97.5%5 / 126
FrontierMath-Tier-4-v2-PrivateMath97.6% ±2.43 / 72
SimpleQA VerifiedKnowledge75.6% ±1.44 / 91
LiveBench MathematicsMath96.8%3 / 66
ScreenSpot ProMultimodal92.7%1 / 20
LiveBench LanguageKnowledge89.4%4 / 66
Vibe Code Bench v1.1Coding89.6% ±2.28 / 123
OSWorld 2.0Agentic72.6%2 / 25
Furniture AssemblyReasoning80.0% ±4.93 / 37
Terminal-Bench 4.0Agentic59.6% ±4.45 / 61
FrontierSWE v2Coding65.5%2 / 22
Medical Long Context Reasoning (MLCR-AA)Reasoning41.7%2 / 22
Agent Arena steerabilityAgentic14.85 / 54
BrowseCompAgentic91.5%6 / 63
Agents' Last ExamAgentic59.3%2 / 21
Artificial Analysis DeepsecBench-AAAgentic36.9%2 / 21
LiveBench Instruction FollowingInstruction75.6%7 / 66
ProofBench v1.1Math99.0% ±1.06 / 54
ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable jobAgentic68.0%2 / 18
FrontierCode 2.0Coding42.8%3 / 27
Agent Arena task outcomeAgentic14.06 / 54
ARC-AGI-3 (semi-private)Reasoning62.7%3 / 26
Bug Hunt BenchCoding45.0%3 / 25
EBR-benchReasoning76.2% ±8.44 / 31
Terminal-Bench 4.0.0Agentic58.2% ±1.43 / 23
ExploitGymAgentic42.4%3 / 23
LiveBench ResearchAgentic85.2%7 / 52
Artificial Analysis GDP.pdfAgentic31.0%3 / 22
Artificial Analysis Terminal-Bench v2.1Agentic88.4%3 / 21
LiveBench Coding 2.0Coding93.7%8 / 52
CWE-bench v1Agentic68.0%2 / 12
Artificial Analysis Long Context ReasoningReasoning80.7%37 / 217
Terminal-Bench 4.1.0Agentic60.3% ±1.44 / 23
DeepSWEAgentic74.1%11 / 59
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic56.3%3 / 16
GDPval-AA normalizedAgentic53.7%25 / 133
Vibe Code Bench 1-100Coding27.6% ±4.15 / 26
PostTrainBench v1.1Coding44.3%3 / 15
Artificial Analysis CWE-Bench-AAAgentic63.3%4 / 19
Artificial Analysis SciCodeCoding56.5%24 / 114
LiveBench DesignAgentic82.0%11 / 52
HealthBench ProfessionalKnowledge64.7%4 / 18
LiveBench Browser UseAgentic84.0%11 / 49
HealthBench HardKnowledge36.6%3 / 13
FrontierCode 1.1 MainCoding53.3%4 / 17
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal69.8%10 / 39
LiveBench CodingCoding80.4%18 / 66
Terminal-Bench-Science 0.1Agentic64.6%3 / 11
Artificial Analysis Terminal-Bench v4.0Agentic59.1%6 / 22
Artificial Analysis AutomationBenchAgentic68.5%7 / 23
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal91.9%8 / 26
MirrorCodeCoding60.0% ±11.812 / 39
Artificial Analysis AnalystAgentAgentic51.2%8 / 26
Multimodal Knowledge Agent BenchmarkAgentic50.2%6 / 19
LiveBench Agentic CodingAgentic57.3%21 / 66
Agent Arena command recoveryAgentic5.516 / 50
Artificial Analysis ITBench-AAAgentic48.6%8 / 24
Humanity's Last Exam with toolsAgentic57.2%13 / 36
CWE-bench v2Agentic64.2%4 / 11
Artificial Analysis Tau3-BankingAgentic41.4%12 / 22
AutomationBenchAgentic41.4%18 / 32
OSWorld 2.1 (all 108 tasks)Agentic75.8%7 / 11
ScarfBench: Cross-Framework Application Migration in Enterprise JavaCoding65.4%7 / 11
Artificial Analysis Cyber IndexAgentic33.4%13 / 19
Artificial Analysis CyberGym-E2E-AAAgentic0.0%17 / 19
Terminal-Bench-ScienceScience62.9% ±5.81 / 44
SRE BenchOther56.9% ±3.11 / 36
MysteryMechanismScience53.2% ±3.41 / 30
ARC-AGI-2 (public eval)Reasoning97.5%5 / 122
ProgramBenchCoding5.5% ±1.65 / 65
SpireClimb (Slay the Spire 2, Ascension 10)Other34.7% ±5.51 / 9
SpireClimbOther34.0% ±4.11 / 9
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other33.3% ±6.11 / 9
CUA-benchOther19.2%1 / 9
ARC-AGI-1 (public eval)Reasoning97.8%14 / 116
BioMysteryBenchScience79.3% ±1.04 / 30
FrontierMath-ErdosMath2.9% ±2.12 / 15
EMBFinance71.7% ±2.511 / 75
MedScribeHealthcare87.9% ±1.916 / 109
Time Horizon Index: KSPOther90.5%2 / 13
MedCodeHealthcare48.5% ±2.129 / 106
Legal Research BenchLegal39.4% ±3.437 / 90
Finance Agent (v2)Finance53.5% ±2.133 / 79
Tax Agent BenchFinance20.7% ±2.952 / 82
HLAB-Vals-1.1Legal7.1% ±1.925 / 35
CyberBench v1.1Other41.1% ±2.645 / 48
OpenAI MRCR v2 8-needle 256K-512KReasoning100.0%—
OpenAI MRCR v2 8-needle 512K-1MReasoning96.3%—
BenchCAD Vision2Code voxel IoU with toolsMultimodal95.9%—
LifeVal: everyday life tasksAgentic80.6%—
Anthropic biomedical-image-analysis evaluationMultimodal77.5%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic73.7%—
GraphWalks BFS 256K–1MReasoning71.8%—
Artificial Analysis HealthBench Professional raw scoreKnowledge69.7%—
HealthBench Professional raw scoreKnowledge68.2%—
BioBasicsBench: everyday wet-lab scienceReasoning67.0%—
Artificial Analysis HealthBench Professional length-adjusted scoreKnowledge66.0%—
FrontierCode 1.1 ExtendedCoding64.5%—
HealthBench length-adjusted scoreKnowledge58.3%—
Artificial Analysis HealthBench length-adjusted scoreKnowledge58.2%—
ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)Multimodal58.0%—
HealthBench raw scoreKnowledge56.9%—
Artificial Analysis HealthBench raw scoreKnowledge56.9%—
AA-Briefcase v1.1, rubric pass rateAgentic52.0%—
PhysicianBenchKnowledge48.0%—
GeneBench-ProReasoning37.8%—
RoboDojo SimAgentic29.8%—
Axiom Bio morphology-to-molecule matchingKnowledge22.8%—
Remote Labor Index (RLI)Agentic20.8%—
Harvey LAB-AA v1.1Agentic8.6%—
STEMBench120: six-domain scientific research progressReasoning2.8%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedEuroEvalMITTerminal-BenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index80.4 → 78.0
23 Oct 202612 Nov 2026
Output / 1M$50.00 → $50.00
23 Oct 202612 Nov 2026

Questions

How much does GPT-6 Astra cost?
$10.00 per 1M input tokens, $50.00 per 1M output tokens.
How good is GPT-6 Astra?
Its index is 78.0 out of 100. That is place 5 of 283 ranked models.
What is GPT-6 Astra best at?
Reasoning: 83.8 out of 100, place 3 of 227.
How fast is GPT-6 Astra?
34 output tokens per second, 3.8 s to the first token.
What is the context window of GPT-6 Astra?
1.05M tokens.
Is GPT-6 Astra open source?
No. Its weights are not published.
When was GPT-6 Astra released?
September 2026.

Badge

GPT-6 Astra on Inference360: place 5 of 283, index 78.0
HTML
<a href="https://inference360.ai/models/gpt-6-astra"><img src="https://inference360.ai/badges/gpt-6-astra.svg" alt="GPT-6 Astra on Inference360: place 5 of 283, index 78.0" height="20"></a>
Markdown
[![GPT-6 Astra on Inference360: place 5 of 283, index 78.0](https://inference360.ai/badges/gpt-6-astra.svg)](https://inference360.ai/models/gpt-6-astra)

Compare with

GPT-6.1 Sol78.0Gemini 4 Argon78.5Claude Opus 5.578.6Muse Wonder Full77.3Claude Sonnet 5.577.2

Lower price

GPT-6.1 Terra77.1 · $5.00/1MGemini 4 Carbon82.7 · $10.00/1MGemini 4 Argon78.5 · $10.00/1M

More from OpenAI

All models
GPT-6 Astra ProUnrankedGPT-Live-1UnrankedGPT Luna LatestUnrankedGPT Terra LatestUnrankedGPT Sol LatestUnranked