Skip to main content

Model profile

GPT-5.6 Luna

OpenAICurrentReleased Jul 9, 2026
Data verified
Overall Score
67.17Public #22 of 200
Arena Elo
Not listed
Eligible category ranks
4of 8
Price (1M tokens)
$1 in / $6 out
API pricing
Speed
Not listed
Context
1M

Evidence coverage

42 of 323 tracked benchmarks are published. 24 are verified and 18 provisional. 6 of 8 categories are measured.

Updated July 23, 2026Methodology
Published / tracked
42 / 323
Verified
24
Provisional
18
Categories with evidence
6 / 8

Evidence by category

  • Agentic15 benchmarks
    Mixed evidence
  • Coding7 benchmarks
    Mixed evidence
  • Reasoning3 benchmarks
    Mixed evidence
  • Knowledge10 benchmarks
    Mixed evidence
  • Math3 benchmarks
    Verified
  • Multilingual0 benchmarks
    Not measured
  • Multimodal3 benchmarks
    Mixed evidence
  • Inst. Following0 benchmarks
    Not measured
ProprietaryReasoning
Confidence:
High
luna

GPT-5.6 Luna ranks #22 out of 200 models on the public leaderboard with an overall score of 67.17/100. It does not yet have enough sourced coverage for BenchLM's verified leaderboard. This places it in the mid-tier of AI models, with strengths in specific benchmark categories.

GPT-5.6 Luna is a proprietary model with a 1M token context window. It uses explicit chain-of-thought reasoning, which typically improves performance on math and complex reasoning tasks at the cost of higher latency and token usage.

GPT-5.6 Luna sits inside the GPT-5.6 family alongside GPT-5.6 Sol, GPT-5.6 Terra. This profile currently has 42 of 323 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Its strongest category is Coding (#5), while its weakest is Multimodal & Grounded (#18). This performance profile makes it particularly well-suited for software development and code generation tasks.

Peer position

Exact provisional scores and ranks for the closest listed peers. A score can appear before a model clears the evidence threshold for a rank, so equal scores can have different rank states.

Range 66.6967.54

  1. Inkling
    Thinking Machines Lab
    #2067.54
    Inkling is #20 with a score of 67.54.
    Compare
  2. Qwen3.7 Plus
    Alibaba
    #2167.22
    Qwen3.7 Plus is #21 with a score of 67.22.
    Compare
  3. GPT-5.6 LunaCurrent model
    OpenAI
    #2267.17
    GPT-5.6 Luna is #22 with a score of 67.17.
  4. GPT-5.2 Pro
    OpenAI
    #2367.01
    GPT-5.2 Pro is #23 with a score of 67.01.
    Compare
  5. GLM-5-Turbo
    Z.AI
    #2466.89
    GLM-5-Turbo is #24 with a score of 66.89.
    Compare
  6. GPT-5.4 nano
    OpenAI
    #2566.79
    GPT-5.4 nano is #25 with a score of 66.79.
    Compare
  7. GPT-5.3 Codex
    OpenAI
    #2666.69
    GPT-5.3 Codex is #26 with a score of 66.69.
    Compare

Category percentile

More

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

  1. Coding97%
    Eligible cohort rank #5 of 122Category score 72.6
  2. Knowledge76%
    Eligible cohort rank #13 of 52Category score 80.9
  3. Agentic87%
    Eligible cohort rank #16 of 119Category score 58.5
  4. Multimodal39%
    Eligible cohort rank #18 of 29Category score 65.9

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #16 of 119Percentile 87thWeight 22%15 benchmarksMixed sources58.5
CodingRank #5 of 122Percentile 97thWeight 20%7 benchmarksMixed sources72.6
ReasoningWeight 17%3 benchmarksMixed sourcesScore pending
KnowledgeRank #13 of 52Percentile 76thWeight 12%10 benchmarksMixed sources80.9
MathRank Not rankedWeight 5%3 benchmarksVerified97.1
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalRank #18 of 29Percentile 39thWeight 12%3 benchmarksMixed sources65.9
Inst. FollowingWeight 5%0 benchmarksNot measuredNot measured

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Agentic15 benchmarks
Terminal-Bench 2.0Provider exact
84.7%Weighted 38%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 84.7% on Terminal-Bench 2.1. BenchLM stores this on the existing Terminal-Bench 2 key.
BrowseCompProvider exact
83.3%Weighted 28%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 83.3% on BrowseComp.
OSWorld 2.0Provider exact
45.6%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 45.6% on OSWorld 2.0.
CyberGymProvider exact
77.9%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 77.9% on CyberGym.
ExploitGymProvider exact
12.4%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 12.4% on ExploitGym.
ToolathlonProvider exact
53.4%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 53.4% on Toolathlon.
AA Agentic IndexReported

Artificial Analysis Agentic Index

45.6%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
GDPval-AAReported

GDPval-AA normalized

54.2%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
GDPval-AAProvider exact
1584Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 1591.8 Elo on GDPval-AA v2.
AA Harvey LABReported

Artificial Analysis Harvey LAB-AA

87.9%Display only
Source: Artificial Analysis: harvey-lab-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA ITBenchReported

Artificial Analysis ITBench-AA

40.3%Display only
Source: Artificial Analysis: itbench-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA Tau3 BankingReported

Artificial Analysis Tau3-Banking

27.2%Display only
Source: Artificial Analysis: tau3-banking leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA AutomationBenchReported

Artificial Analysis AutomationBench

42.2%Display only
Source: Artificial Analysis: automationbench-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
aaTerminalBench21Reported
80.9%Display only
Source: Artificial Analysis: terminalbench-v2-1 leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
APEX-Agents-AAReported
35.8%Display only
Source: Artificial Analysis: apex-agents-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Coding7 benchmarks
SWE-bench ProProvider exact
62.7%Weighted 10%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 62.7% on SWE-Bench Pro.
Terminal-Bench 2.0Provider exact
84.7%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 84.7% on Terminal-Bench 2.1. BenchLM stores this on the existing Terminal-Bench 2 key.
deepSweProvider exact
67.2%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 67.2% on DeepSWE v1.1. BenchLM stores DeepSWE as display-only because the source benchmark is an agent-harness coding signal.
FrontierCode 1.1 ExtendedProvider exact
55.1%Display only
Source: Cognition: GPT-5.6 models in DevinProvenance: Cognition reports GPT-5.6 Luna at 55.1% on the display-only FrontierCode 1.1 Extended subset in Devin.
cursorBench32Benchmark exact
61.1%Display only
Source: Cursor evals: CursorBench 3.2Provenance: Cursor reports GPT-5.6 Luna Max at this exact CursorBench 3.2 score on its public evals page. BenchLM stores it on the GPT-5.6 Luna row as a display-only coding-agent benchmark.
AA Coding IndexReported

Artificial Analysis Coding Index

71.5%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-SciCodeReported

Artificial Analysis SciCode

52.5%Display only
Source: Artificial Analysis: scicode leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Reasoning3 benchmarks
ARC-AGI-3Benchmark exact

Abstraction and Reasoning Corpus for AGI v3

0.2%Display only
Source: ARC Prize: GPT-5.6 verified resultsProvenance: ARC Prize reports GPT-5.6 Luna (Max) at 0.2% on ARC-AGI-3. This is distinct from ARC-AGI-2 and remains display-only.
AA-LCRReported

Artificial Analysis Long Context Reasoning

74.0%Display only
Source: Artificial Analysis: artificial-analysis-long-context-reasoning leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
CritPtSecondary exact

Critical Physics Tasks

20.6%Display only
Source: Artificial Analysis: CritPt Benchmark LeaderboardProvenance: Artificial Analysis reports GPT-5.6 Luna (max) at 20.6% on CritPt.
Knowledge10 benchmarks
GPQAProvider exact

Graduate-Level Google-Proof Q&A

92.3%Weighted 7%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 92.3% on GPQA Diamond. BenchLM stores the value on both gpqa and gpqaDiamond for consistency with provider launch-table mappings.
GPQA-DProvider exact

GPQA Diamond

92.3%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 92.3% on GPQA Diamond.
HealthBench ProfessionalProvider exact
55.7%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 55.7% on HealthBench Professional using the official scoring approach from the HealthBench Professional paper.
HealthBench HardProvider exact
32.0%Display only
Source: OpenAI GPT-5.6 system cardProvenance: OpenAI's July 9, 2026 GPT-5.6 system card reports the length-adjusted HealthBench Hard score for GPT-5.6 Luna. BenchLM stores it as a display-only health benchmark row.
Artificial Analysis Intelligence IndexProvider exact
51.2%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 51.2 on Artificial Analysis Intelligence Index v4.1. BenchLM keeps this as a display-only Artificial Analysis row.
AA-GPQA DiamondReported

Artificial Analysis GPQA Diamond

91.1%Display only
Source: Artificial Analysis: gpqa-diamond leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-HLEReported

Artificial Analysis Humanity's Last Exam

37.2%Display only
Source: Artificial Analysis: humanitys-last-exam leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-Omniscience IndexReported

Artificial Analysis Omniscience Index

-11.2%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-Omniscience AccuracyReported

Artificial Analysis Omniscience Accuracy

41.5%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-Omniscience Hallucination RateReported

Artificial Analysis Omniscience Hallucination Rate

90.1%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
Math3 benchmarks
FrontierMath v2 (Tiers 1-3)Provider exact

FrontierMath v2 Tiers 1-3

78.600%Weighted 30%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 78.6% on FrontierMath v2 Tiers 1-3.
FrontierMath v2 (Tier 4)Provider exact

FrontierMath v2 Tier 4

58.500%Weighted 10%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 58.5% on FrontierMath v2 Tier 4.
FrontierMath (legacy)Provider exact

FrontierMath legacy aggregate

78.6%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 78.6% on FrontierMath v2 Tier 1-3. BenchLM maps the existing FrontierMath field to this Tier 1-3 slice; the separate Tier 4 result is not combined.
Multimodal3 benchmarks
MMMU-ProProvider exact

Massive Multi-discipline Multimodal Understanding Pro

78.4%Weighted 45%
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 78.4% on MMMU Pro without tools.
MMMU-Pro w/ PythonProvider exact

MMMU-Pro with Python

79.5%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 79.5% on MMMU Pro with tools.
AA-MMMU-ProReported

Artificial Analysis MMMU-Pro

78.6%Display only
Source: Artificial Analysis: mmmu-pro leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
External signals1 benchmark
ExploitBenchProvider exact

ExploitBench v8-bench

33%Display only
Source: OpenAI: GPT-5.6Provenance: OpenAI reports GPT-5.6 Luna at 33.2% on ExploitBench. BenchLM stores this as a display-only external security row.

GPT-5.6 Family

luna · luna

Canonical Entry

GPT-5.6 Sol

Frequently Asked Questions

How does GPT-5.6 Luna perform overall in AI benchmarks?

GPT-5.6 Luna currently ranks #22 out of 200 models on BenchLM's provisional leaderboard with an overall score of 67.17. It is created by OpenAI. Its published context window is 1M.

Is GPT-5.6 Luna good for knowledge and understanding?

GPT-5.6 Luna ranks #13 out of 52 models in knowledge and understanding benchmarks with an average score of 80.9. There are stronger options in this category.

Is GPT-5.6 Luna good for coding and programming?

GPT-5.6 Luna ranks #5 out of 122 models in coding and programming benchmarks with an average score of 72.6. It is among the top performers in this category.

Is GPT-5.6 Luna good for mathematics?

GPT-5.6 Luna has visible benchmark coverage in mathematics, but BenchLM does not currently assign it a global category rank there.

Is GPT-5.6 Luna good for reasoning and logic?

GPT-5.6 Luna has visible benchmark coverage in reasoning and logic, but BenchLM does not currently assign it a global category rank there.

Is GPT-5.6 Luna good for agentic tool use and computer tasks?

GPT-5.6 Luna ranks #16 out of 119 models in agentic tool use and computer tasks benchmarks with an average score of 58.5. There are stronger options in this category.

Is GPT-5.6 Luna good for multimodal and grounded tasks?

GPT-5.6 Luna ranks #18 out of 29 models in multimodal and grounded tasks benchmarks with an average score of 65.9. There are stronger options in this category.

Which sibling models are related to GPT-5.6 Luna?

GPT-5.6 Luna belongs to the GPT-5.6 family. Related variants on BenchLM include GPT-5.6 Sol, GPT-5.6 Terra.

Does GPT-5.6 Luna have full benchmark coverage on BenchLM?

Not yet. GPT-5.6 Luna currently has 42 published benchmark scores out of the 323 benchmarks BenchLM tracks. BenchLM only exposes non-generated public benchmark rows, so missing categories stay blank until a sourced evaluation is available.

What is the context window size of GPT-5.6 Luna?

GPT-5.6 Luna has a published context window of 1M, which determines how much text it can process in a single interaction.

Last updated: July 23, 2026 · Runtime metrics stay blank until BenchLM has a sourced snapshot.

Choose with this week’s evidence

Join 2,000+ readers for ranking moves, new releases, pricing changes, and the evidence behind them.

Free. One email per week.