OpenAIUnited States· released Sep 2026

GPT-6 Luna

Name: GPT-6 Luna
Author: OpenAI

Ranked #37, tested so far in 4 of the 5 categories the Index averages.

OpenAI, United States. One of 30 models from this lab on the Index.

Third-party publishedHigh confidence (12+ comparable tests, 4+ Index categories)

Snapshot November 10, 2026

OpenCharts Index

58.8

Ranked #37 (31 to 39) of 80 · 41.2% behind the frontier on average · 22 comparable tests · 4 counting categories

Part of this Index is an estimate. On the 4 categories it is tested in, it averages 60.7. The rest is filled in from models tested in every category, because it has too few results to count there yet: Agentic. Such estimates have been within 2.9 points about nine times in ten.

Its band runs from 55.9 to 61.7: where the Index could land once it is tested in every category, or if any single test were dropped. That covers #31 to #39, and neighbours inside that range are ties.

Needs more results (3 of 5 so far)

Averaging 67.2 on the 3 it has; the score opens at 4 of 5.

Distance to today's frontier on the hardest tests. Not a measure of intelligence.

What it's known for

GPT-6 Luna, in 22 results.

Strongest in math (75.5) and reasoning (56.3). It trails the frontier most in multimodal, 44.6% behind on average. Between 40% and 50% behind the best published results, on average. Part of its Index is an estimate. It has too few results in agentic for that category to count, so it is filled in from models tested in every category. On the categories that do count it averages 60.7. Across its band it could sit anywhere from #31 to #39.

Strongest category

Math

Competition and research mathematics, graded on the final answer or the proof.

75.5

Mock AIME 2024–202598.9% · 99% of best
FrontierMath (Tiers 1–3)78.9% · 83% of best
ProofBench64.0% · 64% of best
FrontierMath Tier 456.1% · 56% of best

Mean of 4 of the category's 4 comparable tests. Counts toward the Index.

Best single result

Mock AIME 2024–2025

Solve competition mathematics problems that have a single exact integer answer.

98.9%

98.9% of the best (100.0%) · #21 of 98.

Olympiad-style problems need long, careful chains of reasoning, and there is no partial credit.

Where it trails

SpireClimb

Win a run of a deck-building strategy game at its hardest setting, playing from screenshots with a mouse and a keyboard.

0.0%

100.0% behind the best published result (34.0%) · #9 of 9.

It tests using a computer the way a person does, and decisions whose cost only shows up much later.

Not measured yet

Covered, but not averaged into the Index:

Knowledgesingle-test category, shown beside the Index · 44.0
Agentic2 of 6 tests · needs 3 · mean 26.3
Human preferencesingle-test category, shown beside the Index · 76.0

Every score, with its source

See it at work

What GPT-6 Luna was asked to do, and how it did.

Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.

Human preference · mean 76.0 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it

Text ArenaLeaderboard

Answer the same prompt as a rival model; real users vote blind on which answer they prefer.

1,441

38% expected win rate against the board leader (1,526)
#74 of 113

max · board 2026-11-09

Source: Arena

Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.

Head to head

GPT-6 Luna against whoever you pick.

Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.

GPT-6 Luna

#37·Index 58.8C-

0 : 21

wins · 1 tie · 22 shared

GPT-6 Astra

#1·Index 93.7A

GPT-6 Astra comes out ahead on 21 of the 22 tests both have taken.

Category means

Reasoning
56.3
96.1
Knowledge
44.0
80.4
Coding
55.5
93.4
Math
75.5
98.8
Agentic
26.3
90.6
Multimodal
55.4
89.6
Human preference
76.0
85.2

Shared tests · biggest gaps first

SpireClimbAgentic
0.0%
34.0%
−100.0
Mystery Game PuzzlesReasoning
7.0%
84.0%
−92.6
WeirdML v3Coding
12.0%
54.0%
−77.8
Furniture AssemblyMultimodal
44.2%
80.0%
−59.7
Chess PuzzlesReasoning
36.0%
74.0%
−51.3
Code Arena (WebDev)Coding
1,578
1,767
−45.2
FrontierMath Tier 4Math
56.1%
97.6%
−41.5
ARC-AGI-2Reasoning
59.3%
95.0%
−37.6
SimpleQA VerifiedKnowledge
41.4%
75.6%
−36.4
ProofBenchMath
64.0%
99.0%
−35.0
CritPtReasoning
19.4%
31.7%
−34.1
GDP.pdfMultimodal
23.0%
34.2%
−32.0
Blueprint-Bench 2Multimodal
31.2%
49.7%
−31.5
APEX-AgentsAgentic
44.3%
64.7%
−24.2
FrontierCodeCoding
42.4%
53.3%
−19.9
FrontierMath (Tiers 1–3)Math
78.9%
93.7%
−15.5
Vision ArenaMultimodal
1,235
1,283
−13.5
ARC-AGI-1Reasoning
86.7%
98.5%
−12.0
Text ArenaHuman preference
1,441
1,474
−9.2
GPQA DiamondReasoning
90.5%
95.8%
−7.4
SciCodeCoding
54.6%
56.5%
−2.7
Mock AIME 2024–2025Math
98.9%
100.0%
tie

Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.

Open GPT-6 Astra

Scores

Every test, every source

The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted, and so is an archived test: one whose publisher no longer covers new models.

Category means

Reasoning56.3 · 6/8
Knowledgebeside44.0 · 1/1
Coding55.5 · 4/6
Math75.5 · 4/4
Agentic26.3 · 2/6
Multimodal55.4 · 4/4
Human preferencebeside76.0 · 1/1
Long context—

Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.

GPQA DiamondEpoch-runReasoning

90.5%

91.7% of the best (96.4%), above chance
58th percentile of 96 · n = 198

max

GPT-6 Luna

GPT-6 Luna, in 22 results.

What GPT-6 Luna was asked to do, and how it did.

GPT-6 Luna against whoever you pick.

Every test, every source

Compare with its neighbours.

The best models, ranked here, working inside Theo.