AIEvals

Models

GPT-6 Luna

Specifications and API prices for GPT-6 Luna, recorded by Artificial Analysis.

OpenAIReleased 22 September 2026Context 1M tokensOutput Not published in the checked sourcePrivate weights$0.10 in / $0.50 out per Mtok

Family GPT-641 benchmarks with a published resultNo first place on the independent columns 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 AI51.22%24of 3325of 41Gemini 4 Argon17.68 pts
Artificial Analysis Intelligence IndexArtificial Analysis3736of 42no list readClaude Opus 5.521 pts
Terminal-Bench 4.0Artificial Analysis12.63%31of 40no list readClaude Sonnet 5.551.01 pts
Vibe Code Bench v1.1Vals AI81.65%17of 3818of 106Claude Sonnet 5.510.74 pts
Terminal-Bench Science 0.1Artificial Analysis8.57%16of 24no list readGPT-6 Astra54.76 pts
ProgramBenchVals AI0.50%13of 2913of 54Claude Opus 5.518.00 pts
MLCR-AAArtificial Analysis16.11%28of 37no list readClaude Sonnet 5.558.89 pts
ARC-AGI-3ARC Prize · Standard0.19%10of 11no list readGPT-6 Astra62.52 pts
AutomationBench-AAArtificial Analysis53.17%28of 40no list readGemini 4 Argon24.34 pts
SciCodeArtificial Analysis54.63%24of 40no list readClaude Opus 5.512.27 pts
GDP.pdfArtificial Analysis22.80%13of 40no list readGPT-6.1 Sol8.20 pts
CritPtArtificial Analysis19.43%23of 42no list readGPT-5.6 Sol12.86 pts
AA-OmniscienceArtificial Analysis0.6536of 42no list readClaude Opus 5.545.77
AA-LCR v1.1Artificial Analysis83.33%11of 42no list readKimi K35.34 pts
Terminal-Bench 2.1Vals AI73.03%21of 3523of 75Claude Opus 5.514.61 pts
ProofBench v1.1Vals AI64.00%16of 3418of 46Claude Sonnet 5.536.00 pts
Harvey's Legal Agent BenchmarkVals AI2.92%29of 3738of 73Muse Spark 1.222.50 pts
MysteryMechanismVals AI19.37%19of 2119of 21GPT-6 Astra33.78 pts
BioMysteryBenchVals AI61.48%17of 1918of 21Claude Sonnet 5.519.63 pts
Code MigrationVals AI42.55%24of 3724of 71Claude Sonnet 5.527.28 pts
CyberBench v1.1Vals AI76.25%4of 314of 43GPT-6 Sol1.73 pts
EMBVals AI68.52%14of 3714of 69Claude Fable 5.18.15 pts
Finance Agent (v2)Vals AI49.87%31of 3738of 73Gemini 4 Argon15.53 pts
IOIVals AI55.56%19of 3119of 38Gemini 4 Argon44.44 pts
Legal Research BenchVals AI30.29%32of 3739of 72Muse Spark 1.3 Max25.00 pts
MedCodeVals AI44.69%23of 3539of 104Claude Opus 518.88 pts
MedScribeVals AI83.71%28of 3641of 106Claude Opus 5.57.72 pts
Public Benefits Bench v1.1Vals AI57.65%27of 2933of 45Claude Opus 519.28 pts
SAGEVals AI48.09%16of 3228of 90Claude Opus 4.78.01 pts
Tax Agent BenchVals AI58.86%29of 3734of 64Claude Fable 5.118.78 pts
GDPval-AA v2.1Artificial Analysis1,438.3830of 42no list readClaude Opus 5.5407.79
AA-Briefcase v1.1Artificial Analysis1,335.5128of 41no list readClaude Opus 5.5486.34
Humanity's Last ExamArtificial Analysis38.51%38of 42no list readClaude Opus 5.522.84 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%
DeepSeek V4.1 FlashDeepSeek51.32%3926.77%84.74%9.05%–
Gemini 3.7 FlashGoogle51.27%3913.64%70.39%–0.00%
GPT-6 LunaOpenAI51.22%3712.63%81.65%8.57%0.50%
Kimi K3Moonshot AI50.70%4412.63%84.96%–2.00%
Hunyuan Hy4 PreviewTencent49.94%––77.48%–0.00%
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 Luna (max)$0.07145.2 tok/s

Results by benchmark

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

BenchmarkIndependent runAs reported by a publisher
OSWorld 2.1 (offline strict pass)Computer useno published result
17.1%OSWorld 2.1 (offline strict pass)Anthropic · 7 October 2026Rank 5 of 6 in this publisher table
HealthBench (length-adjusted)Healthcareno published result
56.3%HealthBench (length-adjusted)Anthropic · 7 October 2026Rank 2 of 2 in this publisher table
HealthBench Professional (length-adjusted)Healthcareno published result
65.5%HealthBench Professional (length-adjusted)Anthropic · 7 October 2026Rank 2 of 4 in this publisher table
PhysicianBenchHealthcareno published result
33.6%PhysicianBenchAnthropic · 7 October 2026Rank 5 of 6 in this publisher table
MLCR-AAMedical reasoning16.11%Artificial Analysisnot published
ARC-AGI-3Interactive reasoning0.19%ARC Prize · Standard
0.59%ARC Prize · Provider adapter
not published
AutomationBench-AAWorkflow automation53.17%Artificial Analysisnot published
SciCodeScientific coding54.63%Artificial Analysisnot published
GDP.pdfDocument reasoning22.80%Artificial Analysisnot published
CritPtScientific reasoning19.43%Artificial Analysisnot published
AA-OmniscienceFactual knowledge0.65Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning83.33%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding12.63%Artificial Analysis
13.64%Vals AI
16.4%Terminal-Bench 4.0Anthropic · 7 October 2026Rank 3 of 4 in this publisher table
16.4%Terminal-Bench 4.0Anthropic · 7 October 2026Rank 5 of 6 in this publisher table
Terminal-Bench 2.1Agentic coding73.03%Vals AInot published
MMMU ProMultimodal reasoning
75.55%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding81.65%Vals AInot published
Terminal-Bench Science 0.1Agentic science8.57%Artificial Analysis
4.29%Vals AI
not published
ProgramBenchAgentic coding0.50%Vals AInot published
ProofBench v1.1Mathematics64.00%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge2.92%Vals AInot published
MysteryMechanismScientific reasoning19.37%Vals AInot published
Vals IndexComposite51.22%Vals AInot published
BioMysteryBenchAgentic science61.48%Vals AInot published
Code MigrationAgentic coding42.55%Vals AInot published
CyberBench v1.1Security76.25%Vals AInot published
EMBFinancial modelling68.52%Vals AInot published
Finance Agent (v2)Financial analysis49.87%Vals AInot published
IOICompetitive programming55.56%Vals AInot published
Legal Research BenchProfessional knowledge30.29%Vals AInot published
MedCodeHealthcare administration44.69%Vals AInot published
MedScribeHealthcare administration83.71%Vals AInot published
Public Benefits Bench v1.1Public services57.65%Vals AInot published
SAGEEducation48.09%Vals AInot published
Tax Agent BenchProfessional knowledge58.86%Vals AInot published
OSWorld 2.1Computer useno published result
48.9%OSWorld 2.1 (offline partial)Anthropic · 7 October 2026Rank 3 of 4 in this publisher table
48.9%OSWorld 2.1 (offline partial)Anthropic · 7 October 2026Rank 5 of 6 in this publisher table
FrontierCode v1.1Agentic codingno published result
42.4%FrontierCode v1.1 (Main)Anthropic · 7 October 2026Rank 3 of 3 in this publisher table
42.4%FrontierCode v1.1 (Main, max)Anthropic · 7 October 2026Rank 3 of 3 in this publisher table
GDPval-AA v2.1Knowledge work1,438.38Artificial Analysis
1,437GDPval-AA v2.1 (max)Anthropic · 7 October 2026Rank 3 of 4 in this publisher table
1,437GDPval-AA v2.1 (max)Anthropic · 7 October 2026Rank 3 of 5 in this publisher table
AA-Briefcase v1.1Knowledge work1,335.51Artificial Analysis
1,336AA-Briefcase v1.1 (max)Anthropic · 7 October 2026Rank 3 of 4 in this publisher table
1,336AA-Briefcase v1.1 (max)Anthropic · 7 October 2026Rank 3 of 5 in this publisher table
ChartographyVisual chart recognitionno published result
29.1%Chartography (no tools)Anthropic · 7 October 2026Rank 3 of 4 in this publisher table
29.1%Chartography (no tools)Anthropic · 7 October 2026Rank 6 of 10 in this publisher table
Humanity's Last ExamMultidisciplinary reasoning38.51%Artificial Analysisnot published
Artificial Analysis Intelligence IndexComposite37Artificial Analysisnot published
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.