Rankings / OpenAI

GPT 5.6 Luna

released 2026-07-09

BenchAtlas Index

as of 2026-07-14

max effort

rank #34

13 families · 5 categories · high

high effort

rank #56

7 families · 5 categories · medium

medium effort

rank #82

7 families · 5 categories · medium

Benchmark evidence

112 results

agentic coding

Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	max effort	53.8 %	independent	LiveBench
Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	xhigh effort	48.8 %	independent	LiveBench
τ²-Bench subset=banking	max effort	27.2 %	independent	Artificial Analysis
τ²-Bench subset=banking	xhigh effort	24.3 %	independent	Artificial Analysis
τ²-Bench subset=banking	high effort	22.3 %	independent	Artificial Analysis
τ²-Bench subset=banking	medium effort	15.3 %	independent	Artificial Analysis
τ²-Bench subset=banking	low effort	12.0 %	independent	Artificial Analysis
τ²-Bench subset=banking	no reasoning	9.1 %	independent	Artificial Analysis

coding

Release 2026-06-25 tasks_counted=2 · livebench_version=2026-06-25	max effort	82.9 %	independent	LiveBench
Release 2026-06-25 tasks_counted=2 · livebench_version=2026-06-25	xhigh effort	76.7 %	independent	LiveBench
SciCode	max effort	52.5 %	independent	Artificial Analysis
SciCode	high effort	50.7 %	independent	Artificial Analysis
SciCode	xhigh effort	50.0 %	independent	Artificial Analysis
SciCode	medium effort	45.8 %	independent	Artificial Analysis
SciCode	low effort	45.6 %	independent	Artificial Analysis
SciCode	no reasoning	39.9 %	independent	Artificial Analysis

data analysis

Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	max effort	78.0 %	independent	LiveBench
Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	xhigh effort	72.9 %	independent	LiveBench

external indices

AA Coding Index	max effort	71.4 points	independent	Artificial Analysis
AA Coding Index	xhigh effort	68.6 points	independent	Artificial Analysis
AA Coding Index	high effort	63.3 points	independent	Artificial Analysis
AA Coding Index	medium effort	50.7 points	independent	Artificial Analysis
AA Coding Index	low effort	44.2 points	independent	Artificial Analysis
AA Coding Index	no reasoning	39.3 points	independent	Artificial Analysis
Intelligence Index v4.1	max effort	51.2 points	independent	Artificial Analysis
Intelligence Index v4.1	xhigh effort	49.1 points	independent	Artificial Analysis
Intelligence Index v4.1	high effort	46.1 points	independent	Artificial Analysis
Intelligence Index v4.1	medium effort	38.1 points	independent	Artificial Analysis
Intelligence Index v4.1	low effort	33.3 points	independent	Artificial Analysis
Intelligence Index v4.1	no reasoning	26.6 points	independent	Artificial Analysis
ECI		155.6 points	independent	Epoch AI Benchmarking Hub
ECI	high effort	155.6 points	independent	Epoch AI Benchmarking Hub
ECI	low effort	155.6 points	independent	Epoch AI Benchmarking Hub
ECI	medium effort	155.6 points	independent	Epoch AI Benchmarking Hub
ECI	max effort	155.6 points	independent	Epoch AI Benchmarking Hub
ECI	xhigh effort	155.6 points	independent	Epoch AI Benchmarking Hub

factuality

SimpleQA Verified

max effort

41.7 %

independent

Epoch AI Benchmarking Hub

2026-07-09

knowledge science

GPQA Diamond	max effort	91.6 %	independent	Epoch AI Benchmarking Hub 2026-07-09
GPQA Diamond implementation=artificial-analysis	max effort	91.1 %	independent	Artificial Analysis
GPQA Diamond implementation=artificial-analysis	xhigh effort	89.5 %	independent	Artificial Analysis
GPQA Diamond implementation=artificial-analysis	high effort	89.2 %	independent	Artificial Analysis
GPQA Diamond implementation=artificial-analysis	medium effort	85.9 %	independent	Artificial Analysis
GPQA Diamond implementation=artificial-analysis	low effort	83.5 %	independent	Artificial Analysis
GPQA Diamond implementation=artificial-analysis	no reasoning	64.5 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	max effort	37.2 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	xhigh effort	35.6 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	high effort	31.6 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	medium effort	24.5 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	low effort	18.8 %	independent	Artificial Analysis
Humanity's Last Exam implementation=artificial-analysis	no reasoning	6.7 %	independent	Artificial Analysis

language

Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	max effort	72.6 %	independent	LiveBench
Release 2026-06-25 tasks_counted=3 · livebench_version=2026-06-25	xhigh effort	70.2 %	independent	LiveBench

long context instruction

AA-LCR	max effort	74.0 %	independent	Artificial Analysis
AA-LCR	xhigh effort	69.7 %	independent	Artificial Analysis
AA-LCR	high effort	69.0 %	independent	Artificial Analysis
AA-LCR	medium effort	66.0 %	independent	Artificial Analysis
AA-LCR	low effort	59.3 %	independent	Artificial Analysis
AA-LCR	no reasoning	36.3 %	independent	Artificial Analysis
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	max effort	60.1 %	independent	LiveBench
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	xhigh effort	57.5 %	independent	LiveBench

multimodal

MMMU

implementation=vals-ai

max effort

85.0 %

independent

Vals AI

professional

CorpFin

max effort

64.2 %

independent

Vals AI

reasoning math

ARC-AGI-1older version split=public_eval · model_type=CoT	max effort	90.8 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	xhigh effort	90.0 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	max effort	88.0 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	xhigh effort	87.7 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	high effort	79.3 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	high effort	76.5 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	medium effort	64.4 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	max effort	60.2 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	max effort	59.5 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	medium effort	56.5 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	xhigh effort	51.4 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	xhigh effort	47.6 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	low effort	47.4 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	low effort	34.2 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	high effort	30.1 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	high effort	29.3 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	medium effort	7.6 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	medium effort	7.4 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	low effort	5.1 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	low effort	2.8 %	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	max effort	0.7 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	max effort	0.7 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	xhigh effort	0.5 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	xhigh effort	0.5 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	max effort	0.3 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	high effort	0.3 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	high effort	0.3 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	xhigh effort	0.2 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	max effort	0.2 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-3older version split=semi_private · model_type=CoT	max effort	0.2 %	independent	ARC Prize Leaderboard
ARC-AGI-3older version split=semi_private · model_type=CoT	medium effort	0.2 %	independent	ARC Prize Leaderboard
ARC-AGI-3older version split=semi_private · model_type=CoT	low effort	0.2 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	high effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	xhigh effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	medium effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	medium effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-3older version split=semi_private · model_type=CoT	high effort	0.1 %	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	high effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=semi_private · model_type=CoT	low effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-2 split=public_eval · model_type=CoT	low effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	medium effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	medium effort	0.1 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=semi_private · model_type=CoT	low effort	0.0 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-1older version split=public_eval · model_type=CoT	low effort	0.0 usd_per_task	independent	ARC Prize Leaderboard
ARC-AGI-3older version split=semi_private · model_type=CoT	xhigh effort	0.0 %	independent	ARC Prize Leaderboard
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	max effort	87.2 %	independent	LiveBench
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	xhigh effort	86.3 %	independent	LiveBench
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	max effort	85.6 %	independent	LiveBench
Release 2026-06-25 tasks_counted=4 · livebench_version=2026-06-25	xhigh effort	84.7 %	independent	LiveBench
OTIS Mock AIME 2024–2025	max effort	98.3 %	independent	Epoch AI Benchmarking Hub 2026-07-09

Agent + model results

systems, not bare-model scores

agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	80.9 %	independent	Artificial Analysis
agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	77.9 %	independent	Artificial Analysis
agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	69.7 %	independent	Artificial Analysis
agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	53.2 %	independent	Artificial Analysis
agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	43.5 %	independent	Artificial Analysis
agent + model Artificial Analysis harness + GPT 5.6 Luna	Terminal-Bench 2.1	39.0 %	independent	Artificial Analysis

These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.