Inference
EN
OpenAI

GPT-6 Luna

GPT-6 Luna is a reasoning model from OpenAI in the GPT-6 family. 57 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 100.65.5±2.1Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.66/sMedian 67/s
First tokenSeconds to the first output token.2.8 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.10batch $0.05Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.50batch $0.25 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1388±4 #173Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-6-luna
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.
$0.01
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.21 counted66.936 / 218
CodingCode writing and repair.9 counted66.832 / 209
ReasoningLogic problems and puzzles.10 counted64.348 / 227
MultimodalTasks with images and text.3 counted65.025 / 84
KnowledgeFacts and expert knowledge.7 counted63.550 / 268
MultilingualTasks in many languages.1 counted53.3—
InstructionTasks with strict prompt rules.1 counted39.8—
MathMath problems.5 counted68.521 / 168

Results

57 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis Long Context ReasoningReasoning83.3%13 / 217
OTIS Mock AIME 2024-2025Math98.9% ±0.722 / 200
LiveBench DesignAgentic82.5%7 / 52
EuroEval ItalianMultilingual62.1% ±0.624 / 174
Critical Physics TasksReasoning19.4%31 / 214
EuroEval SwedishMultilingual61.2% ±0.731 / 195
EuroEval PortugueseMultilingual60.6% ±0.529 / 182
EuroEval FrenchMultilingual62.7% ±0.429 / 178
EuroEval DutchMultilingual59.0% ±0.429 / 175
EuroEval PolishMultilingual56.0% ±0.536 / 209
GPQA diamond · Epoch AIKnowledge90.5% ±1.939 / 217
EuroEval GermanMultilingual50.8% ±0.536 / 191
Artificial Analysis CyberGym-E2E-AAAgenticAbove fitFar above this model's index.77.9%4 / 19
Artificial Analysis Cyber IndexAgentic52.7%4 / 19
Vibe Code Bench v1.1Coding81.6% ±3.426 / 123
EuroEval SpanishMultilingual52.1% ±0.538 / 176
Artificial Analysis MMMU-ProMultimodal79.7%25 / 113
FrontierMath-Tiers-1-3-v2-PrivateMath78.9% ±2.421 / 90
Artificial Analysis Humanity's Last ExamKnowledge38.5%54 / 222
GDPval-AA normalizedAgentic46.6%36 / 133
Artificial Analysis SciCodeCoding54.6%32 / 114
FrontierMath-Tier-4-v2-PrivateMath56.1% ±7.821 / 72
LiveBench Long AgenticAgentic48.1%15 / 51
Terminal-Bench 2.1 · Vals AIAgentic73.0% ±1.723 / 75
Artificial Analysis Omniscience AccuracyKnowledge39.6%54 / 164
ARC-AGI-2 (semi-private)Reasoning59.3%44 / 131
Code MigrationCoding42.6% ±4.427 / 77
Agent Arena steerabilityAgentic3.519 / 54
Agent Arena command recoveryAgentic5.218 / 50
LiveBench CodingCoding79.0%24 / 66
ARC-AGI-1 (semi-private)Reasoning86.7%46 / 126
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal91.0%10 / 26
ProofBench v1.1Math64.0% ±4.822 / 54
LiveBench Browser UseAgentic72.0%20 / 49
Furniture AssemblyReasoning44.2% ±6.416 / 37
HealthBench ProfessionalKnowledge60.8%8 / 18
Medical Long Context Reasoning (MLCR-AA)Reasoning26.1%10 / 22
LiveBench Coding 2.0Coding84.5%24 / 52
IOICoding55.6% ±8.921 / 45
Chess PuzzlesReasoning36.0% ±4.842 / 90
Artificial Analysis CWE-Bench-AAAgentic56.7%9 / 19
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Artificial Analysis DeepsecBench-AAAgentic23.5%11 / 21
SimpleQA VerifiedKnowledge41.4% ±1.649 / 91
DeepSWEAgentic66.6%32 / 59
LiveBench MathematicsMath89.1%37 / 66
LiveBench Data AnalysisReasoning73.4%45 / 80
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal64.1%22 / 39
LiveBench Agentic CodingAgentic51.2%39 / 66
Artificial Analysis GDP.pdfAgentic22.8%13 / 22
Agent Arena task outcomeAgentic−0.432 / 54
HealthBench HardKnowledge31.4%8 / 13
Terminal-Bench 4.0Agentic13.6% ±1.540 / 61
LiveBench ReasoningReasoning81.8%54 / 80
Artificial Analysis AnalystAgentAgentic43.8%19 / 26
Multimodal Knowledge Agent BenchmarkAgentic36.8%14 / 19
Artificial Analysis AutomationBenchAgentic53.2%17 / 23
FrontierCode 2.0Coding21.9%20 / 27
LiveBench LanguageKnowledge73.8%52 / 66
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic33.9%13 / 16
MirrorCodeCoding26.7% ±9.032 / 39
Terminal-Bench 4.0.0Agentic16.4% ±1.419 / 23
Terminal-Bench 4.1.0Agentic15.2% ±1.419 / 23
Mystery Game PuzzlesReasoning7.0% ±2.669 / 80
ExploitGymAgentic11.6%20 / 23
Bug Hunt BenchCoding18.3%22 / 25
OSWorld 2.1 offline subset (82 tasks)Agentic48.9%9 / 10
Artificial Analysis Terminal-Bench v4.0Agentic12.6%20 / 22
ARC-AGI-3 (semi-private)Reasoning0.1%24 / 26
LiveBench Instruction FollowingInstruction55.9%62 / 66
LiveBench ResearchAgentic43.5%49 / 52
CyberBench v1.1Other76.3% ±5.34 / 48
EMBFinance68.5% ±2.816 / 75
ProgramBenchCoding0.5% ±0.519 / 65
ARC-AGI-2 (public eval)Reasoning61.8%36 / 122
ARC-AGI-1 (public eval)Reasoning92.5%40 / 116
MedCodeHealthcare44.7% ±2.340 / 106
MedScribeHealthcare83.7% ±1.942 / 109
Finance Agent (v2)Finance49.9% ±0.245 / 79
Terminal-Bench-ScienceScience4.3% ±2.426 / 44
Legal Research BenchLegal30.3% ±3.254 / 90
SRE BenchOther2.7% ±1.023 / 36
Tax Agent BenchFinance19.9% ±2.955 / 82
SAGEEducation83.4% ±2.321 / 29
Public Benefits Bench v1.1Public benefits57.6% ±1.338 / 50
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other0.0%7 / 9
HLAB-Vals-1.1Legal4.2% ±1.630 / 35
BioMysteryBenchScience61.5% ±2.626 / 30
MysteryMechanismScience19.4% ±2.726 / 30
SpireClimb (Slay the Spire 2, Ascension 10)Other0.0%9 / 9
SpireClimbOther0.0%9 / 9
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic70.5%—
PerceptionBench with toolsMultimodal62.8%—
HealthBench Professional raw scoreKnowledge61.2%—
Agents' Last Exam, Linux subsetAgentic57.1%—
HealthBench length-adjusted scoreKnowledge54.5%—
HealthBench raw scoreKnowledge50.0%—
BioBasicsBench: everyday wet-lab scienceReasoning37.6%—
FrontierCode 2.0, Artificial Analysis evaluationCoding25.2%—
Artificial Analysis AutomationBench, strict pass@1Agentic20.2%—
ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)Multimodal0.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYLiveBenchNo licence statedARC Prize FoundationTerms restrict reuseVals AINo licence statedEuroEvalMITTerminal-BenchNo licence statedLMArenaCC BY 4.0

History

5 releases
Index66.0 → 65.5
23 Oct 202612 Nov 2026
Output / 1M$0.50 → $0.50
23 Oct 202612 Nov 2026

Questions

How much does GPT-6 Luna cost?
$0.10 per 1M input tokens, $0.50 per 1M output tokens.
How good is GPT-6 Luna?
Its index is 65.5 out of 100. That is place 41 of 283 ranked models.
What is GPT-6 Luna best at?
Math: 68.5 out of 100, place 21 of 168.
How fast is GPT-6 Luna?
66 output tokens per second, 2.8 s to the first token.
What is the context window of GPT-6 Luna?
1.05M tokens.
Is GPT-6 Luna open source?
No. Its weights are not published.
When was GPT-6 Luna released?
September 2026.

Badge

GPT-6 Luna on Inference360: place 41 of 283, index 65.5
HTML
<a href="https://inference360.ai/models/gpt-6-luna"><img src="https://inference360.ai/badges/gpt-6-luna.svg" alt="GPT-6 Luna on Inference360: place 41 of 283, index 65.5" height="20"></a>
Markdown
[![GPT-6 Luna on Inference360: place 41 of 283, index 65.5](https://inference360.ai/badges/gpt-6-luna.svg)](https://inference360.ai/models/gpt-6-luna)

Compare with

DeepSeek V4 Pro 081365.5Qwen3.8 Max65.6GPT-5.6 Luna65.5GLM-5.365.8GPT-6 Sol72.8

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from OpenAI

All models
GPT-6 Sol72.8GPT-6 Sol ProUnrankedGPT-6 Luna ProUnrankedGPT-6.1 Sol78.0GPT-6.1 Sol ProUnranked