GPT-5.6 Sol

+ contribute

GPT-5.6 Sol is a closed-weights language model from OpenAI, released July 2026.

81.7±0.5#10 of 719 LLMsfrom 54 benchmark results across 20 sources

$ how it compares

The LLMs ranked either side of GPT-5.6 Sol, out of 828 with a score, then more from OpenAI.

#	model	score	where	released
8	GPT-5.5 Pro OpenAI	82±1	api $67.50/M	2026-04
9	Gemini 4 Argon (limited preview) Google	82±1	limited preview	2026-09
10	GPT-5.6 Sol OpenAI	82±1	api $8.00/M	2026-07
11	Claude Fable 5 Anthropic	81±1	api $20.00/M	2026-06
12	Claude Mythos Preview (limited preview) Anthropic	81±1	limited preview	2026-04
	more from OpenAI
1	GPT-6 Astra OpenAI	87±1	api $20.00/M	2026-09
3	GPT-6.1 Sol OpenAI	85±1	api $4.00/M	2026-09
7	GPT-6 Sol OpenAI	83±1	api $4.00/M	2026-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #11 of 300 · 27% of its score · 14 tests

Beats GPT-5.5 Pro here, though they rank higher overall. Trails Claude Mythos Preview (limited preview), Claude Fable 5 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Terminal-Bench 4.0 +1	37.3% ✎	█▅▅▅▅·▅▃▃▅▅▅ beats 12/20	very hard · avg 0.4%	▇▇ 5.7%	sharp · at its level · several runs
Arena Agent (October 2026 aggregate IPS)	0.0645 IPS ✎	▁▁▁▄▂▃▇█▃▂▁▃ beats 39/49	avg -0.312 IPS	▇▇ 4.4%	sharp
GameLogicBench solve rate (72 Godot game-logic tasks, Claude Code scaffold)	41.7% ✎	█▄·█·▄·██▄·▄ beats 9/11	very hard · avg 4.7%	▇ 2.0%	at its level
SWE-bench Pro (split unspecified) lab	64.6% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 46/55	very hard · avg 5.1%	▇ 1.9%	sharp · at its level
τ³-bench banking_knowledge (v1.0.1 grading)	46.9% ✎	▃▃▂▃·▆▃▂█▅▅▂ beats 22/25	very hard · avg 4.1%	▇ 1.8%	at its level
BrowseComp lab	90.4% ✎	▄··█▄··▄·▄█▄ beats 7/8	hard · avg 32%	▇ 1.8%	typical
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	61.2% ✎	▃··▅··▃█▃▅▃▅ beats 10/12	very hard · avg 3.1%	▇ 1.7%	sharp · at its level
Terminal-Bench 2.1 lab	88.8% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 53/56	hard · avg 21%	▇ 1.7%	typical

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Toolathlon (Tool Decathlon) lab	58.0% ✎	▃··▃··▃··▃█▅ beats 8/8	very hard · avg 13%	▇ 1.5%	at its level
DeepSWE v1.1 (resolved) lab	72.7% ✎	█··▃▃····▃·▃ beats 5/6	hard · avg 21%	▇ 1.2%	at its level
Video2World V2WScore (coding agents reconstruct interactive simulated worlds from videos, OpenCode scaffold)	31.0% ✎	█··▄█·▄····▄ beats 4/6	very hard · avg 6.6%	▇ 1.1%	at its level
Thinkingbox-bench pass@1 (507 stateful business-workflow agent tasks, N=20)	61.9% ✎	▅··▃·▃·█▃▅██ beats 13/15	very hard · avg 4.9%	▇ 1.0%	at its level
KlinikeBench strict pass@1 (333 interactive clinical encounters, Terminus-2)	23.7% ✎	▁▁·▁▁▁▁▅▃█▅█ beats 24/27	very hard · avg 2.4%	▇ 0.8%	typical
EEGAgentBench overall (tool-based EEG analysis agents, 6 tasks)	63.1% ✎	▂·▁▁··▇▇█▂·▃ beats 25/25	hard · avg 29%	▇ 0.6%	blunt · at its level

coding▁▁▁▃▆██▇▆▄▃▁ #19 of 458 · 12% of its score · 6 tests

Trails Grok 4.7, Qwen3.8-Flash-Next, Qwen3.8 Max (0902) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1619.8 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 100/119	avg 981	▇▇ 3.9%	sharp
GameLogicBench solve rate (72 Godot game-logic tasks, Claude Code scaffold)	41.7% ✎	█▄·█·▄·██▄·▄ beats 9/11	very hard · avg 4.7%	▇ 2.0%	at its level
SWE-bench Pro (split unspecified) lab	64.6% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 46/55	very hard · avg 5.1%	▇ 1.9%	sharp · at its level
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	61.2% ✎	▃··▅··▃█▃▅▃▅ beats 10/12	very hard · avg 3.1%	▇ 1.7%	sharp · at its level
DeepSWE v1.1 (resolved) lab	72.7% ✎	█··▃▃····▃·▃ beats 5/6	hard · avg 21%	▇ 1.2%	at its level
Luau Leetcode fully solved (Torpedo Software)	84.6% ✎	▁▁▁▁·▁▂▃▅▃█▇ beats 43/49	hard · avg 18%	▇ 1.0%	typical

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #15 of 393 * · 2.2% of its score · 1 test

Beats GPT-6 Astra, GPT-6.1 Sol, Claude Sonnet 5.5 here, though they rank higher overall. Trails Claude Opus 4.6, Muse Spark 1.1, Claude Opus 4.7 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1483.9 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 363/379	avg 1268	▇ 2.2%	blunt

math▁▂▃▅▆▅▄█▄▁▃▁ #7 of 398 · 32% of its score · 7 tests

Beats Claude Sonnet 5.5, Claude Opus 5, GPT-5.5 Pro here, though they rank higher overall. Trails Claude Fable 5 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tier 4 v2 +1	80.5% ✎	▁▂▁▂█▅▂▃▁▁▁▁ beats 57/63	very hard · avg 0.1%	▇▇▇▇▇▇ 15%	sharp · at its level · several runs
FrontierMath tiers 1–3 v2 +1	89.1% ✎	▂··▁▁▃▄▄█▅▃▃ beats 77/82	very hard · avg 1.2%	▇▇▇▇ 9.0%	sharp · several runs
MathArena ArXivMath June 2026 (48 problems)	88.5% ✎	▂···▄·██▆▆▄▂ beats 17/19	very hard · avg 4.6%	▇▇ 3.9%	sharp
MathArena ArXivLean June 2026 (46 problems)	39.1% ✎	▄····▄▄███·▄ beats 8/9	very hard · avg 0.7%	▇ 2.0%	sharp · at its level
MathArena BrokenArXiv June 2026 (54 problems)	67.3% ✎	▄··█▂▂▂▆▄··▂ beats 12/14	very hard · avg 1.1%	▇ 1.5%	sharp · at its level · noisy
FrontierMath: Erdős problems	0.0% ✎	█··········▆ beats 1/6	very hard · avg 0.0%	▇ 0.3%	sharp · too hard for it · noisy
OTIS mock AIME 2024–2025	100.0% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 191/193	very hard · avg 8.5%	▇ 0.1%	sharp · maxed out · noisy

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #11 of 147 · 4.4% of its score · 6 tests

Beats GPT-6.1 Sol, Claude Fable 5.1, Claude Sonnet 5.5 here, though they rank higher overall. Trails Claude Opus 4.7, Muse Spark 1.2, Qwen3.8-Flash-Next here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena vision (style controlled)	1284.7 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 131/145	avg 1068	▇ 1.5%	blunt
MMMU Pro lab	83.0% ✎	▅·····▃·▃·█▃ beats 7/7	hard · avg 40%	▇ 1.2%	typical
Arena document (style controlled)	1488.7 ✎	▁▁▃▃▄▃▇█▅▇▄▅ beats 32/37	avg 1335	▇ 0.7%	blunt
SciDocBench scientific-document workflows (overall, 496 instances)	61.0% ✎	▃▃···█▃▅▃·▃▅ beats 10/11	hard · avg 26%	▇ 0.7%	blunt · at its level
ScreenSpot-Pro lab	76.9% ✎	▂▃▅▆██▅▃···▂ beats 23/24	medium · avg 48%	▇ 0.2%	blunt · small run
PureDocBench document parsing (Avg3 over Clean/Digital/Real)	69.2% ✎	▂·▁·▂▂▅█·▁▂▂ beats 12/22	medium · avg 62%	▇ 0.0%	blunt · at its level

web search▂▁▂▆▄█▃▃▁▁·▁ #7 of 80 * · 2.7% of its score · 2 tests

Beats GPT-6 Astra, Claude Opus 5.5, Claude Fable 5.1 here, though they rank higher overall. Trails GPT-4o (2024-08-06), GPT-4.1 mini, GPT-4.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp lab	90.4% ✎	▄··█▄··▄·▄█▄ beats 7/8	hard · avg 32%	▇ 1.8%	typical
Arena search systems (style controlled)	1255.8 ✎	▁·▁▁▂▂▂█▅▂▁▃ beats 30/30	avg 1055	▇ 0.9%	blunt

other tests20 not in a task · 28% of its score

test	result	vs the others in its run	difficulty	counts	why
EBR-bench (Epoch)	44.8% ✎	▄▂·▆█▂▆▄▆▆·▄ beats 17/23	very hard · avg 0.3%	▇▇▇ 6.3%	sharp · at its level
Mystery Game Puzzles (Epoch)	58.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 68/74	very hard · avg 9.4%	▇▇ 4.4%	sharp · at its level · noisy
Furniture Assembly (Epoch)	56.7% ✎	█▆▆▆▆▂▅▂▂▂·▅ beats 23/30	very hard · avg 1.2%	▇▇ 4.1%	sharp · at its level
Chess Puzzles (Epoch)	64.0% ✎	█▁▁▁▁▂▁▄▂▂▁▃ beats 134/135	very hard · avg 5.9%	▇ 2.3%	sharp · at its level · noisy
KMMLU-HARD (Daekeun Kim, zero-shot)	85.0% ✎	█···▁▁▁▂▃·▆▃ beats 22/23	very hard · avg 29%	▇ 1.7%	sharp
KoBALT-700 (Daekeun Kim)	84.3% ✎	▄▄█·····▄··▄ beats 4/5	hard · avg 37%	▇ 1.0%	typical
SimpleQA Verified	69.7% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 71/78	very hard · avg 8.0%	▇ 0.8%	at its level
KMMLU-Pro (Daekeun Kim)	93.6% ✎	▄·▄█·····▄·▄ beats 4/5	medium · avg 40%	▇ 0.8%	typical

12 more tests

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	93.5% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 198/208	hard · avg 40%	▇ 0.8%	sharp · maxed out
AutomationBench public set (Zapier README)	45.8% ✎	▄······▄█··▄ beats 1/4	very hard · avg 10%	▇ 0.8%	at its level
CLIcK Korean (Daekeun Kim)	95.5% ✎	▆▄▄▄█▄▂·▄█·▄ beats 22/23	medium · avg 58%	▇ 0.7%	maxed out
Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite)	65.3% ✎	▁▁▃▂▂▄▁▃▄▄▄█ beats 31/38	very hard · avg 14%	▇ 0.7%	at its level
MuSR Korean (Daekeun Kim)	85.3% ✎	█·····███·██ beats 4/5	medium · avg 42%	▇ 0.6%	typical
MirrorCode (Epoch)	20.0% ✎	█·▄▄·▄·▄·▄·█ beats 3/8	very hard · avg 1.5%	▇ 0.5%	sharp · at its level · noisy
HealthBench Hard lab	31.1% ✎	▂··▂····▂▄·█ beats 3/8	very hard · avg 9.7%	▇ 0.5%	blunt · at its level
HAE-RAE 1.0 (Daekeun Kim)	94.6% ✎	▄▄▄▂▂█▂▄▂█▂▄ beats 21/22	medium · avg 63%	▇ 0.5%	maxed out
HealthBench lab	55.6% ✎	▃·▃·····█▃▃▅ beats 3/8	hard · avg 24%	▇ 0.5%	blunt · at its level
RobloxQA v2.0 multiple choice (Torpedo Software)	97.9% ✎	▁▁▃▁█▄▃▃▆▂▃▂ beats 45/49	easy · avg 78%	▇ 0.4%	maxed out
NTCIR SciClaimEval Subtask 1 pair-accuracy (dev, single model)	70.2% ✎	▃▃···▃·▃▃█·▃ beats 7/8	medium · avg 45%	▇ 0.3%	blunt · at its level
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	87.6% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 19/98	very easy · avg 90%	▇ 0.0%	blunt

$ where to run it

Through an API

api · $8.00/M · $4 in / $20 out per 1M tokens · 1.1M context · platform.openai.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

$ how the score was worked out · Epoch AI — internal evaluatio… 33%, OpenAI vendor-reported benchm… 24%, LMArena leaderboard dataset 14% and 17 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Epoch AI — internal evaluations	11	▇▇▇▇▇▇▇▇▇▇ 33%	CC BY 4.0
OpenAI vendor-reported benchmark scores for its own models (openai.com announcement posts and Deployment Safety Hub system cards) · lab-reported	12	▇▇▇▇▇▇▇ 24%	Factual vendor-reported score (not licensed content)
LMArena leaderboard dataset	6	▇▇▇▇ 14%	CC BY 4.0
MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets	3	▇▇ 7%	CC BY-SA 4.0
Daekeun Kim Korean LLM benchmark runs	6	▇▇ 5%	MIT
Terminal-Bench 4.0 official leaderboard submissions	1	▇ 3%	Apache-2.0
GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions (arXiv:2609.21562v2)	1	▇ 2%	CC BY 4.0
τ²/τ³-bench leaderboard: Sierra's own evaluation runs	1	▇ 2%	MIT
E2E-SWE: Benchmarking LLMs on Building Working Codebases from Scratch (arXiv:2609.38335v1)	1	▇ 2%	CC BY 4.0
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode	2	▇ 1%	MIT
Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos (arXiv:2610.04432v2)	1	▇ 1%	CC BY-SA 4.0
One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows (arXiv:2608.19741v4)	1	▇ 1%	CC BY 4.0
KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy (arXiv:2609.38480v2)	1	▇ 1%	CC BY 4.0
Zapier AutomationBench README: public-set pass rates	1	▇ 1%	MIT
Miðeind Icelandic LLM Leaderboard	1	▇ 1%	CC BY 4.0
SciDocBench: A Workflow-Centered Benchmark and Data Pipeline for Scientific Document Understanding (arXiv:2609.05141v2)	1	▇ 1%	CC BY 4.0
EEGAgentBench: Benchmarking LLM Agents on Short- and Long-Horizon EEG Analysis (arXiv:2609.31632v1)	1	▇ 1%	CC BY 4.0
SciTrue: Reliable Scientific Claim Validation with Frontier and Open Language Models at the NTCIR SciClaimEval Task — authors' own single-model dev evaluation	1	▇ 0%	CC BY 4.0
How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)	1	▇ 0%	CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 81.7