Inference
EN
OpenAI

GPT-5.6 Luna

GPT-5.6 Luna is a reasoning model from OpenAI in the GPT-5.6 family. 61 benchmarks count toward its index, in 8 categories.

  • Closed weights
  • Reasoning
  • Input: file · image · text
  • Released 2026-07
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.51/sMedian 67/s
First tokenSeconds to the first output token.1.7 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.20batch $0.10Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.20batch $0.60 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1431±4 #94Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-5.6-luna
Max outputMaximum output tokens per request.
128K
Knowledge cutoffLast date of training data.
16 Feb 2026
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.02
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.19 counted64.747 / 218
CodingCode writing and repair.15 counted67.429 / 209
ReasoningLogic problems and puzzles.9 counted66.341 / 227
MultimodalTasks with images and text.3 counted65.823 / 84
KnowledgeFacts and expert knowledge.8 counted62.155 / 268
MultilingualTasks in many languages.1 counted53.1—
InstructionTasks with strict prompt rules.1 counted48.2—
MathMath problems.5 counted68.422 / 168

Results

61 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis Long Context ReasoningReasoning83.7%10 / 217
EuroEval FrenchMultilingual68.9% ±0.410 / 178
IOI v1Coding72.9% ±5.14 / 61
EuroEval SwedishMultilingual66.9% ±0.515 / 195
SWE-benchCoding93.0% ±1.19 / 87
LiveBench CodingCoding82.9%7 / 66
EuroEval SpanishMultilingual57.9% ±0.419 / 176
Graduate-Level Google-Proof Q&AKnowledge92.3%18 / 141
GPQA Diamond · Vals AIKnowledge91.7% ±1.718 / 135
Critical Physics TasksReasoning20.6%29 / 214
Artificial Analysis GPQA DiamondKnowledge91.1%27 / 199
OTIS Mock AIME 2024-2025Math98.3% ±1.228 / 200
EuroEval GermanMultilingual52.6% ±0.627 / 191
GPQA diamond · Epoch AIKnowledge91.6% ±1.731 / 217
Terminal-Bench 2.1 · Vals AIAgentic79.0% ±1.011 / 75
EuroEval PortugueseMultilingual60.7% ±0.427 / 182
EuroEval DutchMultilingual59.3% ±0.628 / 175
EuroEval ItalianMultilingual60.0% ±0.530 / 174
FrontierMath-Tiers-1-3-v2-PrivateMath82.1% ±2.318 / 90
SkillsBenchCoding60.4% ±4.87 / 35
Artificial Analysis Humanity's Last ExamKnowledge39.5%49 / 222
GDPval-AA normalizedAgentic48.0%30 / 133
GPQA Diamond · BenchLMKnowledge92.3%19 / 84
APEX-Agents-AAAgentic35.8%7 / 30
EuroEval PolishMultilingual54.2% ±0.450 / 209
Terminal-Bench 2.1 · BenchLMAgentic84.7%21 / 86
Artificial Analysis Omniscience AccuracyKnowledge42.7%42 / 164
FrontierMath-Tier-4-v2-PrivateMath61.0% ±7.719 / 72
MMMU ProMultimodal85.0% ±0.925 / 92
SWE-bench ProCoding62.7%23 / 83
Code MigrationCoding44.5% ±4.222 / 77
Artificial Analysis MMMU-ProMultimodal78.6%33 / 113
Vibe Code Bench 1-100Coding22.6% ±4.08 / 26
Vibe Code Bench v1.1Coding77.1% ±3.139 / 123
ToolathlonAgentic53.4%7 / 22
ARC-AGI-2 (semi-private)Reasoning59.5%43 / 131
ARC-AGI-1 (semi-private)Reasoning88.0%42 / 126
Artificial Analysis SciCodeCoding53.6%38 / 114
Chess PuzzlesReasoning43.0% ±5.031 / 90
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal91.2%9 / 26
LiveBench Data AnalysisReasoning78.0%28 / 80
IOICoding61.8% ±11.616 / 45
MMLU ProKnowledge86.0% ±0.350 / 135
BrowseCompAgentic83.3%24 / 63
LiveBench ReasoningReasoning85.6%37 / 80
ProofBench v1.1Math60.0% ±4.925 / 54
LiveBench Browser UseAgentic68.0%23 / 49
Massive Multi-discipline Multimodal Understanding ProMultimodal78.4%25 / 53
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Furniture AssemblyReasoning42.5% ±6.418 / 37
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal66.9%19 / 39
LiveBench DesignAgentic76.0%26 / 52
DeepSWEAgentic67.2%30 / 59
OSWorld 2.0Agentic45.6%13 / 25
Mystery Game PuzzlesReasoning21.0% ±4.143 / 80
HealthBench HardKnowledge32.0%7 / 13
SimpleQA VerifiedKnowledge41.0% ±1.651 / 91
LiveBench Long AgenticAgentic30.4%30 / 51
cursorBench32Coding61.1%11 / 18
Agent Arena command recoveryAgentic0.531 / 50
VulcanBench v3Coding85.5%9 / 14
LiveBench MathematicsMath87.2%43 / 66
CyberGymAgentic77.9%20 / 29
Terminal-Bench 4.0Agentic11.6% ±1.043 / 61
Artificial Analysis ITBench-AAAgentic40.3%17 / 24
LiveBench Coding 2.0Coding73.6%37 / 52
LiveBench Agentic CodingAgentic48.4%47 / 66
Terminal-Bench 3.0Agentic14.3%10 / 13
Terminal-Bench 4.1.0Agentic17.6% ±1.518 / 23
Terminal-Bench 4.0.0Agentic17.3% ±1.518 / 23
ARC-AGI-3 (semi-private)Reasoning0.2%21 / 26
ExploitGymAgentic12.4%19 / 23
ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable jobAgentic7.0%15 / 18
LiveBench LanguageKnowledge72.6%56 / 66
LiveBench Instruction FollowingInstruction60.1%56 / 66
Artificial Analysis Terminal-Bench v2.1Agentic80.9%18 / 21
HealthBench ProfessionalKnowledge55.7%16 / 18
LiveBench ResearchAgentic51.0%47 / 52
cursorBench40Coding35.9%19 / 21
MirrorCodeCoding21.1% ±9.836 / 39
TaxEval v2Finance76.2% ±0.86 / 143
CyberBench v1.1Other73.6% ±5.611 / 48
EMBFinance67.1% ±2.918 / 75
MortgageTaxFinance67.3% ±0.926 / 97
LegalBenchLegal84.0% ±0.443 / 149
CorpFin v2Finance64.2% ±0.938 / 131
Finance Agent (v2)Finance55.0% ±0.324 / 79
ARC-AGI-2 (public eval)Reasoning60.2%38 / 122
MedScribeHealthcare84.4% ±2.636 / 109
ARC-AGI-1 (public eval)Reasoning90.8%43 / 116
MedCodeHealthcare42.4% ±2.351 / 106
Legal Research BenchLegal36.5% ±3.346 / 90
ProgramBenchCoding0.0%34 / 65
Tax Agent BenchFinance24.8% ±3.146 / 82
Public Benefits Bench v1.1Public benefits61.2% ±1.331 / 50
SRE BenchOther1.5% ±0.827 / 36
BioMysteryBenchScience61.5% ±0.726 / 30
Terminal-Bench-ScienceScience0.0%41 / 44
MysteryMechanismScience14.4% ±2.429 / 30
MMMU-Pro with PythonMultimodal79.5%—
FrontierCode 1.1 ExtendedCoding55.1%—
Artificial Analysis EnterpriseOps-GymAgentic40.8%—

Sources

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

History

5 releases
Index67.0 → 65.5
23 Oct 202612 Nov 2026
Output / 1M$1.20 → $1.20
23 Oct 202612 Nov 2026

Questions

How much does GPT-5.6 Luna cost?
$0.20 per 1M input tokens, $1.20 per 1M output tokens.
How good is GPT-5.6 Luna?
Its index is 65.5 out of 100. That is place 41 of 283 ranked models.
What is GPT-5.6 Luna best at?
Math: 68.4 out of 100, place 22 of 168.
How fast is GPT-5.6 Luna?
51 output tokens per second, 1.7 s to the first token.
What is the context window of GPT-5.6 Luna?
1.05M tokens.
Is GPT-5.6 Luna open source?
No. Its weights are not published.
When was GPT-5.6 Luna released?
July 2026.

Badge

GPT-5.6 Luna on Inference360: place 41 of 283, index 65.5
HTML
<a href="https://inference360.ai/models/gpt-5-6-luna"><img src="https://inference360.ai/badges/gpt-5-6-luna.svg" alt="GPT-5.6 Luna on Inference360: place 41 of 283, index 65.5" height="20"></a>
Markdown
[![GPT-5.6 Luna on Inference360: place 41 of 283, index 65.5](https://inference360.ai/badges/gpt-5-6-luna.svg)](https://inference360.ai/models/gpt-5-6-luna)

Compare with

DeepSeek V4 Pro 081365.5GPT-6 Luna65.5Muse Spark Open65.1Muse Spark 1.264.9dots3-note Preview63.4

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-5.6 Sol73.0GPT-5.6 Terra68.9GPT-5.6 Sol ProUnrankedGPT-5.6 Terra ProUnrankedGPT-5.6 Luna ProUnranked