GPT-5.6 Luna

Name: GPT-5.6 Luna
Author: OpenAI

Multimodal

OpenAI

GPT-5.6 Luna is the fastest and most cost-efficient tier of OpenAI's GPT-5.6 family, designed for cost-sensitive, high-volume workloads. It brings strong capability at OpenAI's lowest GPT-5.6 price, supports max reasoning effort, and has a 1.05M-token context window.

Key Specifications

Parameters

Context

1.1M

Release Date

July 9, 2026

Average Score

52.3%

API Documentation Research Paper Results Blog

Timeline

Key dates in the model's history

Announcement

July 9, 2026

Today

August 27, 2026

Technical Specifications

Parameters

Training Tokens

Knowledge Cutoff

February 16, 2026

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$0.20

Output (per 1M tokens)

$1.20

Max Input Tokens

1.1M

Max Output Tokens

128.0K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis

GPQA

GPQA Diamond. Max reasoning effort. • Self-reported

92.3%

Other Tests

Specialized benchmarks

Agents' Last Exam

Agents' Last Exam. Max reasoning effort. • Self-reported

50.3%

ARC-AGI-3

ARC-AGI-3. Max reasoning effort. • Self-reported

0.2%

Artificial Analysis

Artificial Analysis Intelligence Index v4.1 (max). Score 51. • Self-reported

51.0%

AutomationBench

AutomationBench. Max reasoning effort. • Self-reported

14.9%

BenchCAD

BenchCAD (no tools). Max reasoning effort. • Self-reported

63.1%

BenchCAD (with Python tool)

BenchCAD with Python tool. Max reasoning effort. • Self-reported

73.9%

Big Finance Bench

Big Finance Bench. Max reasoning effort. • Self-reported

36.0%

BrowseComp

BrowseComp. Max reasoning effort. • Self-reported

83.3%

Capture-the-Flag Challenges (Internal)

Capture-the-Flag Challenges (Internal). Max reasoning effort. • Self-reported

85.2%

Connectors

Connectors production benchmark pass rate. • Self-reported

99.9%

DeepSWE

DeepSWE v1.1. Max reasoning effort. • Self-reported

67.2%

DeepSWE 1.1

DeepSWE v1.1 leaderboard, mini-swe-agent harness, max effort; Pass@1 67% ± 4%. • Self-reported

67.0%

ExploitBench

ExploitBench. Max reasoning effort. • Self-reported

33.2%

ExploitGym

ExploitGym, six-hour cap. Max reasoning effort. • Self-reported

12.4%

FrontierCode 1.1

FrontierCode 1.1 current leaderboard; mergeability score at max effort: 39.8%. • Self-reported

39.8%

FrontierMath

FrontierMath Tier 1-3 (v2). Max reasoning effort. • Self-reported

78.6%

FrontierMath Tier 4 (v2)

FrontierMath Tier 4 (v2). Max reasoning effort. • Self-reported

58.5%

GDP.pdf

gdp.pdf. Max reasoning effort. • Self-reported

22.7%

GeneBench-Pro

GeneBench Pro. Max reasoning effort. • Self-reported

10.8%

Graphwalks BFS >128k

GraphWalks BFS 256k f1. Max reasoning effort. • Self-reported

81.3%

Graphwalks BFS 1M

GraphWalks BFS 1M f1. Max reasoning effort. • Self-reported

51.2%

HealthBench

HealthBench, length-adjusted score (unadjusted 55.4). • Self-reported

55.8%

HealthBench Consensus

HealthBench Consensus, length-adjusted score (unadjusted 95.1). • Self-reported

95.1%

HealthBench Hard

HealthBench Hard, length-adjusted score (unadjusted 31.4). • Self-reported

32.0%

HealthBench Professional

HealthBench Professional, length-adjusted score (unadjusted 59.8). • Self-reported

55.7%

Internal Research Debugging Evaluation

Internal Research Debugging Evaluation. Max reasoning effort. • Self-reported

50.8%

KernelGen 1P

KernelGen 1P. Max reasoning effort. • Self-reported

22.4%

LifeSciBench

LifeSciBench. Max reasoning effort. • Self-reported

51.2%

Management Consulting Tasks (Internal)

Management Consulting Tasks (Internal). Max reasoning effort. • Self-reported

35.4%

MedChemBench (Internal)

MedChemBench (Internal). Max reasoning effort. • Self-reported

30.4%

MMMU-Pro

MMMU Pro (no tools). Max reasoning effort. • Self-reported

78.4%

MMMU-Pro (with tools)

MMMU Pro (with tools). Max reasoning effort. • Self-reported

79.5%

MRCR v2 (8-needle)

OpenAI MRCR v2 8-needle, 256K-512K. Max reasoning effort. • Self-reported

41.3%

MRCR v2 (8-needle, 512K-1M)

OpenAI MRCR v2 8-needle, 512K-1M. Max reasoning effort. • Self-reported

41.3%

NanoGPT

NanoGPT. Max reasoning effort. • Self-reported

1.7%

OSWorld 2.0

OSWorld 2.0 binary completion. Max reasoning effort. • Self-reported

45.6%

PostTrainBench Lite

PostTrainBench Lite mean reward. Max reasoning effort. • Self-reported

29.6%

RSI Index

RSI Index (aggregate self-improvement). Max reasoning effort. • Self-reported

41.9%

Search and Function-Calling

Search and Function-Calling production benchmark pass rate. • Self-reported

89.7%

SEC-bench Pro

SEC-Bench Pro. Max reasoning effort. • Self-reported

48.9%

SWE-Bench Pro

SWE-Bench Pro. Max reasoning effort. • Self-reported

62.7%

Terminal-Bench 2.1

Terminal-Bench 2.1. Max reasoning effort. • Self-reported

84.7%

Toolathlon

Toolathlon. Max reasoning effort. • Self-reported

53.4%

License & Metadata

License

proprietary

Announcement Date

July 9, 2026

Last Updated

August 27, 2026

Compare GPT-5.6 Luna

All comparisons

vs GPT-5.6 Terra vs GPT-5.6 Sol vs GPT-5.1 Thinking vs GPT-5 High vs GPT-5.4 vs GPT-5.1 High vs GPT-5.1 Codex High vs GPT-5.2

Similar Models

All Models

GPT-5

OpenAI

Best score:0.9 (HumanEval)

Released:Aug 2025

Price:$1.25/1M tokens

GPT-5.5

OpenAI

Best score:1.0 (TAU)

Released:Apr 2026

Price:$5.00/1M tokens

GPT-5.6 Terra

OpenAI

Best score:0.9 (GPQA)

Released:Jul 2026

Price:$2.00/1M tokens

GPT-5.1 Instant

OpenAI

Best score:1.0 (TAU)

Released:Nov 2025

Price:$0.30/1M tokens

GPT-5.4 mini

OpenAI

Best score:0.9 (TAU)

Released:Mar 2026

Price:$0.42/1M tokens

GPT-5.4 nano

OpenAI

Best score:0.9 (TAU)

Released:Mar 2026

Price:$0.12/1M tokens

GPT-5.5 Instant

OpenAI

Best score:0.9 (GPQA)

Released:May 2026

Price:$5.00/1M tokens

GPT-5.6 Sol

OpenAI

Best score:0.9 (GPQA)

Released:Jul 2026

Price:$5.00/1M tokens

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.