OpenAIUnited States· released Jul 2026

GPT-5.6 Sol

Name: GPT-5.6 Sol
Author: OpenAI

Best published result on 2 of the 20 tests it has taken.

OpenAI, United States. One of 26 models from this lab on the Index.

Third-party publishedMedium confidence (7+ comparable tests, 3+ Index categories)

Available in Theo as Theo Explore

Theo orchestrates it for the steps it does best and always shows which engine answered.

Use it in Theo

Snapshot September 8, 2026

OpenCharts Index

89.8

Ranked #5 of 36 · 10.2% behind the frontier on average · 20 comparable tests · 3 counting categories

Drop any single test and the Index lands between 88.7 and 92.4.

Needs more results (3 of 8 so far)

Averaging 92.5 on the 3 it has; the score opens at 4 of 8.

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

What it's known for

GPT-5.6 Sol, in 20 results.

Strongest in reasoning (94.5) and math (90.8). It holds the best published result on CritPt and Mock AIME 2024–2025. It trails the frontier most in coding, 16.0% behind on average. No comparable result yet in long context. Between 10% and 15% behind the best published results, on average.

Best published result onCritPt32.3%Mock AIME 2024–2025100.0%

Strongest category

Reasoning

Hard, novel problems: graduate science, abstract puzzles, expert exams.

94.5

CritPt32.3% · best
ARC-AGI-197.5% · 99% of best
ARC-AGI-292.5% · 97% of best
GPQA Diamond93.5% · 97% of best
SimpleBench64.8% · 79% of best

Mean of 5 of the category's 6 comparable tests. Counts toward the Index.

Best single result

CritPt

Solve an unpublished research-level physics problem to a numeric or symbolic answer checked by an official server.

32.3%

Best published result of 78 models.

Research physics is far past textbook recall. Only a few models produce anything a physicist would accept.

Where it trails

Code Arena (WebDev)

Build a web app from the same prompt as a rival model; real users vote blind on the result.

1,617

47.5 points short of parity with the board leader (1,797) · #10 of 86.

People judging finished apps side by side is the most honest measure of front-end quality.

Not measured yet

No comparable result yet in this category. A blank is a blank, never a zero, and it does not lower the Index.

Long context

Covered, but not averaged into the Index:

Knowledgesingle-test category, shown beside the Index · 92.2
Agentic3 of 8 tests · needs 4 · mean 88.8
Multimodalsingle-test category, shown beside the Index · 91.2
Human preferencesingle-test category, shown beside the Index · 93.1

Every score, with its source

See it at work

What GPT-5.6 Sol 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.

Reasoning · mean 94.5 across 5 comparable tests · counts toward the Index

CritPtLeaderboard

Solve an unpublished research-level physics problem to a numeric or symbolic answer checked by an official server.

32.3%

best published result
of 78 models

max

Source: CritPt / Artificial Analysis

ARC-AGI-1Leaderboard

Solve the original Abstraction and Reasoning Corpus: small grids, a handful of examples, one hidden rule.

97.5%

99.0% of the best (98.5%)
#4 of 51 · n = 100

xhigh

Close to saturated at the frontier; scores depend on the compute budget a lab chose.

Source: ARC Prize Foundation

ARC-AGI-2Hard setLeaderboard

Infer the hidden rule from a few input and output grid pairs, then apply it to a new grid.

92.5%

97.4% of the best (95.0%)
#2 of 51 · n = 120

max

Scores depend on the compute budget a lab chose; ARC Prize publishes cost per task beside every score and this ranking does not.

Source: ARC Prize Foundation

GPQA DiamondEpoch-run

Answer a PhD-level multiple-choice question in physics, chemistry or biology that a web search will not settle.

93.5%

96.8% of the best (95.8%), above chance
#8 of 78 · n = 198

max

Frontier models cluster in the eighties and nineties; Epoch's standard errors on this test are 2 to 3 points.

Source: Epoch AI (internal runs)

SimpleBenchLeaderboard

Answer everyday trick questions about the physical and social world that most people find easy.

64.8%

79.1% of the best (81.9%)
#15 of 46

Source: SimpleBench / LM Council

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-5.6 Sol 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-5.6 Sol

#5·Index 89.8A-

0 : 10

wins · 4 ties · 14 shared

GPT-6 Astra

#1·Index 99.0A+

GPT-6 Astra comes out ahead on 10 of the 14 tests both have taken.

Category means

Reasoning
94.5
99.6
Knowledge
92.2
100.0
Coding
84.0
97.7
Math
90.8
99.8
Agentic
88.8
98.5
Multimodal
91.2
—
Human preference
93.1
—

Shared tests · biggest gaps first

Code Arena (WebDev)Coding
1,617
1,797
−47.5
ProofBenchMath
83.0%
99.0%
−16.0
FrontierMath Tier 4Math
82.9%
97.6%
−15.0
APEX-AgentsAgentic
39.9%
46.7%
−14.3
FrontierCodeCoding
47.5%
53.3%
−10.8
SimpleQA VerifiedKnowledge
69.7%
75.6%
−7.8
FrontierMath (Tiers 1–3)Math
89.1%
93.7%
−4.9
WeirdMLCoding
88.8%
92.9%
−4.5
GPQA DiamondReasoning
93.5%
95.8%
−3.2
ARC-AGI-2Reasoning
92.5%
95.0%
−2.6
CritPtReasoning
32.3%
31.7%
tie
ARC-AGI-1Reasoning
97.5%
98.5%
tie
SciCodeCoding
56.9%
56.5%
tie
Mock AIME 2024–2025Math
100.0%
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.

Category means

Reasoning94.5 · 5/6
Knowledgebeside92.2 · 1/1
Coding84.0 · 5/7
Math90.8 · 4/5
Agentic88.8 · 3/8
Multimodalbeside91.2 · 1/1
Human preferencebeside93.1 · 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

93.5%

96.8% of the best (95.8%), above chance
91th percentile of 78 · n = 198

max

GPT-5.6 Sol

GPT-5.6 Sol, in 20 results.

What GPT-5.6 Sol was asked to do, and how it did.

GPT-5.6 Sol against whoever you pick.

Every test, every source

Compare with its neighbours.

The best models, ranked here, working inside Theo.