OpenAI

GPT Luna 5.6 High

Version 5.6-high benchmark runs across the shared Superbash visual prompts.

Official benchmark profile

How GPT Luna 5.6 High scores beyond our visual tests.

OpenAI describes Luna as the fastest and most cost-efficient GPT-5.6 variant. These rows use the shared GPT-5.6 comparison table so scores line up against Sol, Terra, GPT-5.5, Claude, and Gemini where published.

OpenAI sourceJuly 2026Source report →

Coding

SWE-bench Pro

62.7%

Agentic coding

Terminal-Bench 2.1

84.7%

Computer use

OSWorld 2.0

45.6%

Full official benchmark table9 rows with source settings and peer charts

Benchmark	Area	Score	Setting / comparison
SWE-bench Pro	Coding	62.7%	Shared OpenAI GPT-5.6 comparison table. Claude Fable 580.0% Claude Opus 4.869.2% GPT-5.6 Sol64.6% GPT-5.6 Terra63.4% GPT Luna 5.6 High62.7% GPT-5.559.4% Gemini 3.1 Pro54.2%
Terminal-Bench 2.1	Agentic coding	84.7%	Shared OpenAI GPT-5.6 comparison table. GPT-5.6 Sol88.8% GPT-5.6 Terra87.4% GPT-5.585.6% GPT Luna 5.6 High84.7% Claude Fable 583.1% Claude Opus 4.878.9% Gemini 3.1 Pro70.7%
OSWorld 2.0	Computer use	45.6%	High-effort computer-use comparison table. GPT-5.6 Sol62.6% Claude Opus 4.854.8% GPT-5.6 Terra50.2% GPT-5.547.5% GPT Luna 5.6 High45.6%
BrowseComp	Tool use	83.3%	High-effort browsing-agent comparison table. GPT-5.6 Sol Ultra92.2% GPT-5.6 Sol90.4% Claude Mythos 588.0% Claude Mythos Preview87.9% GPT-5.6 Terra87.5% Gemini 3.1 Pro85.9% GPT-5.584.4% Claude Opus 4.884.3%
BenchCAD	Computer-aided design	63.1%	Vision2Code score without Python tool. GPT-5.6 Sol70.6% GPT Luna 5.6 High63.1% GPT-5.6 Terra62.3% GPT-5.544.4% Claude Mythos 538.4% Claude Mythos Preview35.5% Claude Opus 4.827.3%
BenchCAD with Python tool	Tool use	73.9%	Vision2Code score with Python tool. GPT-5.6 Sol83.4% GPT-5.6 Terra78.2% GPT Luna 5.6 High73.9% GPT-5.559.4% Claude Mythos 556.7% Claude Mythos Preview56.4% Claude Opus 4.848.1%
GPQA Diamond	Academic reasoning	92.3%	Shared reasoning benchmark table. GPT-5.6 Sol94.6% Gemini 3.1 Pro94.3% Claude Mythos 594.1% GPT-5.593.6% GPT-5.6 Terra92.9% Claude Fable 592.6% GPT Luna 5.6 High92.3% Claude Opus 4.892.0%
FrontierMath Tier 1-3 v2	Math	78.6%	Shared reasoning benchmark table. GPT-5.6 Sol89.0% Claude Fable 587.0% GPT-5.585.3% GPT-5.6 Terra84.9% Claude Opus 4.880.0% GPT Luna 5.6 High78.6% Gemini 3.1 Pro59.6%
FrontierMath Tier 4 v2	Math	58.5%	Shared reasoning benchmark table. Claude Fable 587.8% GPT-5.6 Sol83.0% GPT-5.572.5% GPT-5.6 Terra68.3% GPT Luna 5.6 High58.5% Claude Opus 4.856.1%

Visual prompt runs

Benchmark runs

Open each generated scene, or compare the same prompt across models.

9 runs

Helm's Deep

Fortress siege scene testing scale, lighting, architecture, and cinematic atmosphere.

Open run →Compare models →

Hogwarts Broom Flight Simulator

Broom-flight scene testing depth, motion cues, castle scale, and fantasy mood.

Open run →Compare models →

Jabberwock

Dark fantasy encounter testing creature design, forest mood, and narrative staging.

Open run →Compare models →

Low Poly World

Stylized island build testing composition, color, and low-poly worldbuilding.

Open run →Compare models →

Office Life

Workplace vignette testing everyday scene logic, objects, and believable office detail.

Open run →Compare models →

Petri Dish

Microscopic ecosystem testing organic forms, scientific clarity, and cellular detail.

Open run →Compare models →

Universe Simulator

Cosmic system testing orbital structure, glowing bodies, scale, and simulation readability.

Open run →Compare models →

Vice City

Neon coastal city testing vehicles, architecture, atmosphere, and dense urban layout.

Open run →Compare models →

Yingzao Fashi Assembly

Timber assembly scene testing structure, joinery, construction order, and material clarity.

Open run →Compare models →