AIEvals

Models

GPT-6 Astra

OpenAI’s GPT-6 flagship for complex coding, research and agentic work.

OpenAIReleased 3 September 2026Context 1M tokensOutput 128k tokensPrivate weights$10.00 in / $50.00 out per Mtok

Family GPT-661 benchmarks with a published result8 independent first places among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI63.13%6of 336of 41Gemini 4 Argon5.77 pts
Artificial Analysis Intelligence IndexArtificial Analysis533of 42no list readClaude Opus 5.55 pts
Terminal-Bench 4.0Artificial Analysis59.09%3of 40no list readClaude Sonnet 5.54.55 pts
Vibe Code Bench v1.1Vals AI89.59%6of 386of 106Claude Sonnet 5.52.80 pts
Terminal-Bench Science 0.1Artificial Analysis63.33%1of 24no list readGPT-6 Astra—
ProgramBenchVals AI5.50%3of 293of 54Claude Opus 5.513.00 pts
τ³-BankingArtificial Analysis41.44%15of 30no list readQwen3.8 Max9.90 pts
ITBench-AAArtificial Analysis48.59%7of 18no list readGPT-5.6 Sol7.62 pts
AA-AnalystAgentArtificial Analysis51.25%4of 18no list readGemini 3.7 Flash8.75 pts
MLCR-AAArtificial Analysis35.00%12of 37no list readClaude Sonnet 5.540.00 pts
ARC-AGI-3ARC Prize · Standard62.71%1of 11no list readGPT-6 Astra—
SWE-Bench Pro V2 FullScale AI / SEAL96.90%4of 10no list readClaude Opus 52.50 pts
SWE-Bench Pro V2 HARDScale AI / SEAL90.20%3of 10no list readClaude Opus 57.80 pts
Vals RSI IndexVals AI27.07%6of 206of 23Claude Opus 5.510.24 pts
Vibe Code Bench 1–100Vals AI27.64%4of 204of 20Claude Opus 5.52.72 pts
CUA-benchVals AI19.17%1of 81of 8GPT-6 Astra—
AutomationBench-AAArtificial Analysis68.49%5of 40no list readGemini 4 Argon9.02 pts
SciCodeArtificial Analysis56.48%18of 40no list readClaude Opus 5.510.42 pts
GDP.pdfArtificial Analysis31.00%1of 40no list readGPT-6.1 Sol—
CritPtArtificial Analysis31.71%2of 42no list readGPT-5.6 Sol0.58 pts
AA-OmniscienceArtificial Analysis43.43of 42no list readClaude Opus 5.53.02
AA-LCR v1.1Artificial Analysis80.67%25of 42no list readKimi K38.00 pts
Terminal-Bench 2.1Vals AI87.27%2of 352of 75Claude Opus 5.50.37 pts
ProofBench v1.1Vals AI99.00%4of 345of 46Claude Sonnet 5.51.00 pts
Harvey's Legal Agent BenchmarkVals AI5.42%23of 3729of 73Muse Spark 1.220.00 pts
MysteryMechanismVals AI53.15%1of 211of 21GPT-6 Astra—
BioMysteryBenchVals AI79.26%3of 193of 21Claude Sonnet 5.51.85 pts
Code MigrationVals AI67.74%3of 373of 71Claude Sonnet 5.52.09 pts
CyberBench v1.1Vals AI41.07%29of 3140of 43GPT-6 Sol36.91 pts
EMBVals AI71.70%9of 379of 69Claude Fable 5.14.97 pts
Finance Agent (v2)Vals AI53.54%23of 3728of 73Gemini 4 Argon11.86 pts
IOIVals AI100.00%1of 311of 38Gemini 4 Argon—
Legal Research BenchVals AI39.42%24of 3726of 72Muse Spark 1.3 Max15.87 pts
MedCodeVals AI48.49%16of 3527of 104Claude Opus 515.08 pts
MedScribeVals AI87.91%12of 3614of 106Claude Opus 5.53.52 pts
SAGEVals AI46.37%21of 3236of 90Claude Opus 4.79.73 pts
SRE BenchVals AI56.87%1of 151of 16GPT-6 Astra—
Tax Agent BenchVals AI63.34%21of 3724of 64Claude Fable 5.114.30 pts
DeepSWE v1.1Datacurve74.12%1of 201of 28GPT-6 Astra—
GDPval-AA v2.1Artificial Analysis1,541.8923of 42no list readClaude Opus 5.5304.28
AA-Briefcase v1.1Artificial Analysis1,568.8610of 41no list readClaude Opus 5.5252.99
Humanity's Last ExamArtificial Analysis54.68%7of 42no list readClaude Opus 5.56.67 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%Vibe Code Bench v1.1%Terminal-Bench Science 0.1%ProgramBench%
Claude Fable 5.1Anthropic65.83%5352.02%90.26%43.33%7.00%
Claude Opus 5Anthropic63.67%5148.99%88.40%28.57%3.00%
GPT-6 AstraOpenAI63.13%5359.09%89.59%63.33%5.50%
Claude Fable 5Anthropic61.39%5042.42%90.35%–2.00%
GPT-6.1 SolOpenAI61.15%5256.06%88.93%58.10%–
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

Cost and speed, as published

Additional measurements from Artificial Analysis’ own row. Read 30 September 2026.

As publishedCost per taskOutput speed
GPT-6 Astra (max)$3.2655.7 tok/s

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
τ³-BankingTool use41.44%Artificial Analysisnot published
ITBench-AAIT operations48.59%Artificial Analysisnot published
AA-AnalystAgentData analysis51.25%Artificial Analysisnot published
MLCR-AAMedical reasoning35.00%Artificial Analysisnot published
ARC-AGI-3Interactive reasoning62.71%ARC Prize · Standard
99.95%ARC Prize · Provider adapter
not published
SWE-Bench Pro V2 FullRepository coding96.90%Scale AI / SEALnot published
SWE-Bench Pro V2 HARDRepository coding90.20%Scale AI / SEALnot published
Vals RSI IndexAutonomous research27.07%Vals AInot published
Vibe Code Bench 1–100Application development27.64%Vals AInot published
CUA-benchComputer use19.17%Vals AInot published
PLawBenchProfessional knowledgeno published result
69.1%PLawBenchZ.ai · 12 October 2026Rank 5 of 5 in this publisher table
LABBench 2Scientific reasoningno published result
85.4%LABBench 2Google DeepMind · October 2026Rank 2 of 4 in this publisher table
RiemannBenchMathematicsno published result
72.0%RiemannBenchGoogle DeepMind · October 2026Rank 2 of 4 in this publisher table
GraphWalks (up to 128k)Long contextno published result
98.7%GraphWalks (Up to 128k, BFS F1)Google DeepMind · October 2026Rank 2 of 4 in this publisher table
GraphWalks (256k to 1M)Long contextno published result
71.8%GraphWalks (256k to 1M, BFS F1)Google DeepMind · October 2026Rank 2 of 4 in this publisher table
LVBenchVideo understandingno published result
87.5%LVBench (no tools)Google DeepMind · October 2026Rank 2 of 4 in this publisher table
CWE-bench v1Securityno published result
68.0%CWE-bench v1Google DeepMind · October 2026Rank 1 of 4 in this publisher table
AutomationBench-AAWorkflow automation68.49%Artificial Analysisnot published
SciCodeScientific coding56.48%Artificial Analysisnot published
GDP.pdfDocument reasoning31.00%Artificial Analysisnot published
CritPtScientific reasoning31.71%Artificial Analysisnot published
AA-OmniscienceFactual knowledge43.4Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning80.67%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding59.09%Artificial Analysis
59.60%Vals AI
58.2%Terminal-Bench 4.0Google DeepMind · October 2026Rank 2 of 4 in this publisher table
57.9%Terminal-Bench 4.0OpenAI · 3 September 2026Rank 1 of 6 in this publisher table
57.9%Terminal-Bench 4.0Anthropic · 22 September 2026Rank 2 of 5 in this publisher table
Terminal-Bench 2.1Agentic coding87.27%Vals AI
88.39%Artificial Analysis
not published
GPQA DiamondScience reasoning
96.06%Artificial Analysis
96.0%GPQA DiamondOpenAI · 3 September 2026Rank 1 of 6 in this publisher table
MMMU ProMultimodal reasoning
86.88%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding89.59%Vals AI
89.6%Vibe Code BenchGoogle DeepMind · October 2026Rank 4 of 4 in this publisher table
Terminal-Bench Science 0.1Agentic science63.33%Artificial Analysis
62.86%Vals AI
68.1%Terminal-Bench Science 0.1Google DeepMind · October 2026Rank 1 of 4 in this publisher table
64.6%Terminal-Bench Science 0.1OpenAI · 3 September 2026Rank 1 of 5 in this publisher table
64.6%Terminal-Bench-Science 0.1Anthropic · 22 September 2026Rank 1 of 5 in this publisher table
68.1%Terminal-Bench Science 0.1OpenAI · 29 September 2026Rank 1 of 4 in this publisher table
ProgramBenchAgentic coding5.50%Vals AInot published
ProofBench v1.1Mathematics99.00%Vals AInot published
LegalBenchProfessional knowledgeno published result
87.1%LegalBenchZ.ai · 12 October 2026Rank 2 of 5 in this publisher table
Harvey's Legal Agent BenchmarkProfessional knowledge5.42%Vals AI
5.4%Harvey's Legal Agent BenchmarkZ.ai · 12 October 2026Rank 5 of 5 in this publisher table
5.4%Harvey's Legal Agent BenchmarkGoogle DeepMind · October 2026Rank 3 of 4 in this publisher table
MysteryMechanismScientific reasoning53.15%Vals AInot published
Vals IndexComposite63.13%Vals AI
63.1%Vals IndexGoogle DeepMind · October 2026Rank 4 of 4 in this publisher table
BioMysteryBenchAgentic science79.26%Vals AInot published
Code MigrationAgentic coding67.74%Vals AInot published
CyberBench v1.1Security41.07%Vals AInot published
EMBFinancial modelling71.70%Vals AInot published
Finance Agent (v2)Financial analysis53.54%Vals AI
53.5%Vals Finance Agent v2Google DeepMind · October 2026Rank 4 of 4 in this publisher table
IOICompetitive programming100.00%Vals AInot published
Legal Research BenchProfessional knowledge39.42%Vals AInot published
MedCodeHealthcare administration48.49%Vals AInot published
MedScribeHealthcare administration87.91%Vals AInot published
SAGEEducation46.37%Vals AInot published
SRE BenchSecurity56.87%Vals AInot published
Tax Agent BenchProfessional knowledge63.34%Vals AInot published
OSWorld 2.0Computer useno published result
72.6%OSWorld 2.0 (offline set, partial reward)Google DeepMind · October 2026Rank 1 of 2 in this publisher table
72.6%OSWorld 2.0 (offline set, partial reward)OpenAI · 3 September 2026Rank 1 of 3 in this publisher table
73.5%OSWorld 2.0 (offline set, partial reward)OpenAI · 29 September 2026Rank 1 of 3 in this publisher table
AutomationBenchBusiness workflowsno published result
41.4%AutomationBenchGoogle DeepMind · October 2026Rank 3 of 4 in this publisher table
41.4%AutomationBench 1.0.6OpenAI · 3 September 2026Rank 1 of 5 in this publisher table
41.4%AutomationBenchAnthropic · 22 September 2026Rank 1 of 5 in this publisher table
41.4%AutomationBench 1.0.6OpenAI · 29 September 2026Rank 2 of 5 in this publisher table
DeepSWE v1.1Software engineering74.12%Datacurve
74.1%DeepSWE v1.1Google DeepMind · October 2026Rank 3 of 4 in this publisher table
74.1%DeepSWE v1.1OpenAI · 3 September 2026Rank 1 of 6 in this publisher table
74.1%DeepSWE v1.1OpenAI · 29 September 2026Rank 2 of 3 in this publisher table
FrontierCode v1.1Agentic codingno published result
53.3FrontierCode 1.1 MainOpenAI · 3 September 2026Rank 3 of 6 in this publisher table
53.3%FrontierCode v1.1 (Main)Anthropic · 22 September 2026Rank 2 of 5 in this publisher table
GDPval-AA v2.1Knowledge work1,541.89Artificial Analysis
1,542GDPval-AA v2.1Anthropic · 22 September 2026Rank 5 of 5 in this publisher table
AA-Briefcase v1.1Knowledge work1,568.86Artificial Analysisnot published
ChartographyVisual chart recognitionno published result
71.0%Chartography (no tools)Google DeepMind · October 2026Rank 2 of 4 in this publisher table
Humanity's Last ExamMultidisciplinary reasoning54.68%Artificial Analysis
57.2%Humanity's Last Exam (with tools)OpenAI · 3 September 2026Rank 4 of 4 in this publisher table
57.2%Humanity's Last Exam (with tools)Anthropic · 22 September 2026Rank 4 of 4 in this publisher table
ARC-AGI-2Abstract reasoningno published result
95.0%ARC-AGI-2OpenAI · 3 September 2026Rank 1 of 5 in this publisher table
FrontierMath Tier 4 (v2)Mathematicsno published result
97.6%FrontierMath Tier 4 (v2)OpenAI · 3 September 2026Rank 1 of 5 in this publisher table
Agents' Last ExamComputer useno published result
34.2%Agent's Last Exam (binary pass rate)Google DeepMind · October 2026Rank 3 of 3 in this publisher table
59.3%Agents' Last ExamOpenAI · 3 September 2026Rank 1 of 4 in this publisher table
Artificial Analysis Intelligence IndexComposite53Artificial Analysisnot published
FrontierSWEAgentic researchno published result
65.5%FrontierSWE v2Google DeepMind · October 2026Rank 1 of 4 in this publisher table
PostTrainBenchAI researchno published result
44.3%PostTrainBench v1.1Google DeepMind · October 2026Rank 3 of 4 in this publisher table
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Sources

Read on 17 October 2026.