Model

GPT-5.6 Sol

CLOSED

OpenAI · GPT-5.6 family · released Jul 9, 2026

Flagship (Sol) tier of the three-tier GPT-5.6 family, billed as OpenAI's best coding model yet (Terminal-Bench 2.1 SOTA); limited-preview June 26 under US govt review, GA July 9, 2026. Sibling tiers GPT-5.6 Terra and GPT-5.6 Luna are tracked as their own corpus models in the same family (OpenAI: 'durable capability tiers that can advance on their own cadence').

ReasoningCodingVisionFunction callingTool useAgentic

2915.3

Elo · rank #12

Parameters

Undisclosed

Active params

—

Context

1.05M tokens

Architecture

Undisclosed architecture (presumed dense transformer); flagship tier of the three durable capability tiers Sol/Terra/Luna, each tracked as its own corpus model (like Claude Opus/Sonnet/Haiku)

License

Proprietary

Languages

—

API price (in/out)

$5 / $30

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AA-BriefcaseAgents1505#5

best: Claude Opus 5 · 1720

Agents' Last ExamAgents52.7%#2

best: GPT-6 Astra · 59.3%

ARC-AGI-1Reasoning96.5%#4

best: GPT-6 Astra · 98.5%

ARC-AGI-2Reasoning92.5%#2

best: GPT-6 Astra · 95.0%

ARC-AGI-3Reasoning7.8%#2

best: Claude Opus 5 · 30.2%

AutomationBenchAgents18.1%#9

best: Qwen3.8-Max-0902 · 50.8%

BenchCAD (Vision2Code)Vision83.3%#3

best: GPT-6 Astra · 95.9%

BrowseCompAgents90.4%#3

best: Kimi K3 · 91.2%

CursorBenchCoding67.2%#5

best: Claude Fable 5.1 · 73.4%

CyberGymCoding84.5%#2

best: GLM-5.3 · 84.5%

DeepSearchQAAgents93.0%#1

best: this model · 93.0%

DeepSWECoding72.7%#3

best: Muse Spark 1.3 · 75.4%

ExploitBenchAgents76.5%#2

best: GPT-6 Astra · 100.0%

FrontierBenchAgents34.4%#2

best: Claude Opus 5 · 43.3%

FrontierCode (Extended)Coding60.6%#6

best: Claude Fable 5 · 64.9%

FrontierCodeCoding47.5%#5

best: Claude Fable 5 · 53.5%

FrontierMath (Tiers 1-3)Math89.0%#1

best: this model · 89.0%

FrontierMath (Tier 4)Math83.0%#1

best: this model · 83.0%

GDP.PDFVision40.0%#1

best: this model · 40.0%

GDPval-AAAgents1747.8#8

best: Claude Fable 5 · 1932

GPQA DiamondReasoning94.6%#2

best: GPT-6 Astra · 96.0%

Harvey Legal Agent BenchmarkAgents2.5%#7

best: Grok 4.6 · 15.8%

HLE-VerifiedKnowledge54.5%#2

best: Gemini 3.8 Flash · 54.9%

Humanity's Last ExamReasoning47.2%#12

best: Claude Opus 5 · 64.7%

IFBenchReasoning72.7%#25

best: MiniMax M3 · 83.0%

JobBenchAgents45.4%#6

best: Claude Opus 5 · 65.7%

LVBenchVision82.1%#3

best: Gemini 3.8 Flash · 87.8%

MMMU-ProVision83.0%#4

best: Claude Opus 4.7 · 85.5%

MRCR v2 8-needle (512K-1M)Reasoning73.8%#3

best: Muse Spark 1.3 · 98.1%

MRCR v2 8-needle (256K-512K)Reasoning91.5%#3

best: GPT-6 Astra · 100.0%

OSWorld 2.0Agents62.6%#5

best: Claude Fable 5.1 · 77.9%

ScreenSpot-ProVision76.9%#3

best: GPT-6 Astra · 92.7%

SRE-BenchCoding55.9%#2

best: GPT-6 Astra · 88.0%

SWE-Atlas Codebase QnACoding53.5%#2

best: Muse Spark 1.3 · 59.4%

SWE-bench ProCoding64.6%#7

best: Claude Fable 5.1 · 81.2%

τ³-BankingAgents33.0%#1

best: this model · 33.0%

Terminal-Bench 2.0Coding88.8%#2

best: Gemini 3.8 Flash · 89.4%

Terminal-Bench 3.0Agents34.6%#1

best: this model · 34.6%

Terminal-Bench 4.0Agents37.3%#6

best: Claude Mythos 5.1 · 60.9%

Terminal-Bench Science 0.1Agents22.4%#5

best: GPT-6 Astra · 64.6%

Toolathlon (Verified)Agents74.9%#4

best: GLM-5.3-Flash · 78.4%

Vals Finance AgentAgents53.8%#4

best: Gemini 3.8 Flash · 61.4%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$5

Output / M tok

$30

GPT-5.6 family

Elo progression across releases

GPT-5.6 LunaJul 20262651.5 GPT-5.6 SolJul 20262915.3 GPT-5.6 TerraJul 20262688.0 GPT-5.6 CyberAug 20261000.0

Resources

DOCSdevelopers.openai.com/api/docs/models/gpt-5.6-sol APIAvailable via OpenAI API

API price $5/$30 · each benchmark row carries its own source badge (see methodology)