GPT-5.5

+ contribute

GPT-5.5 is a closed-weights language model from OpenAI, released April 2026.

80.1±0.5#14 of 719 LLMsfrom 68 benchmark results across 21 sources

$ how it compares

The LLMs ranked either side of GPT-5.5, out of 828 with a score, then more from OpenAI.

#	model	score	where	RAM	released
12	Claude Mythos Preview (limited preview) Anthropic	81±1	limited preview		2026-04
13	GPT-5.4 Pro OpenAI	81±1	api $67.50/M		2026-03
14	GPT-5.5 OpenAI	80±1	api $11.25/M		2026-04
15	MiMo-V2.6-Pro Xiaomi	79±1	api $0.54/M +local	~443GB	2026-09
16	Hy4 preview (770B-A49B) Tencent	79±1	api $1.25/M +local	~248GB	2026-08
	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▁▁▃▆█▆▅▅▅▄▃▁ #25 of 300 · 26% of its score · 22 tests

Trails GLM-5.3-Flash, Grok 4.7, Qwen3.8 Max (0902) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena Agent (October 2026 aggregate IPS)	0.0415 IPS ✎	▁▁▁▄▂▃▇█▂▂▁▃ beats 36/49	avg -0.312 IPS	▇▇ 3.8%	sharp
Agents' Last Exam overall full-pass rate (fixed OpenClaw + GUI-as-Tool)	21.1% ✎	▄···█▄▄▄█▄·█ beats 10/10	very hard · avg 1.7%	▇ 2.3%	typical
SWE-bench Pro (split unspecified) lab	59.4% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 39/55	very hard · avg 5.1%	▇ 1.7%	sharp · at its level
BrowseComp lab	84.4% ✎	▄··█▄··▄·▄█▄ beats 3/8	hard · avg 32%	▇ 1.7%	typical
K-BrowseComp-Verified (Korean web-browsing agent questions, pass@1)	45.7% ✎	▃····█··▃█·▃ beats 8/8	very hard · avg 3.4%	▇ 1.7%	at its level
Terminal-Bench 2.1 lab	85.6% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 48/56	hard · avg 21%	▇ 1.6%	typical
τ³-bench banking_knowledge (v1.0.1 grading)	44.6% ✎	▃▃▂▃·▆▃▂█▅▅▂ beats 21/25	very hard · avg 4.1%	▇ 1.6%	at its level
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	46.1% ✎	▃··▅··▃█▃▅▃▅ beats 7/12	very hard · avg 3.1%	▇ 1.5%	sharp · at its level

14 more tests

test	result	vs the others in its run	difficulty	counts	why
Toolathlon (Tool Decathlon) lab	55.6% ✎	▃··▃··▃··▃█▅ beats 7/8	very hard · avg 13%	▇ 1.3%	at its level
Humanity's Last Exam with tools lab	52.2% ✎	▃·▃·▅····▅·█ beats 5/8	very hard · avg 12%	▇ 1.2%	at its level
Terminal-Bench 2.0 lab	82.7% ✎	██···██··███ beats 6/6	very hard · avg 14%	▇ 1.0%	typical
DeepSWE v1.1 (resolved) lab	67.0% ✎	█··▃▃····▃·▃ beats 1/6	hard · avg 21%	▇ 1.0%	at its level
MCP-Atlas (public set)	75.3% ✎	█··█·██·████ beats 4/7	hard · avg 29%	▇ 0.9%	typical
Legal Research Bench (agentic, all-pass)	40.0% ✎	▄█··▄▄▄██·▄█ beats 11/12	very hard · avg 5.8%	▇ 0.8%	at its level
Agents' Last Exam (metric/harness unspecified) lab	46.9% ✎	▂·▂·█▂···▃▄▄ beats 7/11	very hard · avg 11%	▇ 0.7%	at its level
KlinikeBench strict pass@1 (333 interactive clinical encounters, Terminus-2)	16.8% ✎	▁▁·▁▁▁▁▅▃█▅█ beats 16/27	very hard · avg 2.4%	▇ 0.7%	typical
OSWorld-Verified lab	78.7% ✎	█▄·····▄·▄▄▄ beats 6/6	hard · avg 17%	▇ 0.7%	typical
RWSearch	85.3% ✎	████··█····█ beats 5/5	medium · avg 44%	▇ 0.6%	typical
EEGAgentBench overall (tool-based EEG analysis agents, 6 tasks)	50.0% ✎	▂·▁▁··▇▇█▂·▁ beats 19/25	hard · avg 29%	▇ 0.5%	blunt · at its level
SWE-bench Pro lab	58.6% ✎	▄·····▄·▄▄▄█ beats 6/6	hard · avg 28%	▇ 0.5%	blunt · at its level
TerminalWorld-Verified (200 tasks) with Terminus-2	53.5% ✎	▄··▄·▄·█···▄ beats 1/5	hard · avg 27%	▇ 0.4%	blunt · at its level
τ²-bench Telecom lab	98.0% ✎	▃·▅▅▃▃··█▃▃▅ beats 12/13	hard · avg 16%	▇ 0.3%	sharp · maxed out · noisy

coding▁▁▁▃▆██▇▆▄▃▁ #46 of 458 · 9.0% of its score · 6 tests

Trails Mistral Large 4 (public preview), Claude Sonnet 4.6, Qwen3.8 27B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1512.3 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 73/119	avg 981	▇ 3.3%	sharp
SWE-bench Pro (split unspecified) lab	59.4% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 39/55	very hard · avg 5.1%	▇ 1.7%	sharp · at its level
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	46.1% ✎	▃··▅··▃█▃▅▃▅ beats 7/12	very hard · avg 3.1%	▇ 1.5%	sharp · at its level
DeepSWE v1.1 (resolved) lab	67.0% ✎	█··▃▃····▃·▃ beats 1/6	hard · avg 21%	▇ 1.0%	at its level
Luau Leetcode fully solved (Torpedo Software)	85.1% ✎	▁▁▁▁·▁▂▃▅▃█▇ beats 45/49	hard · avg 18%	▇ 1.0%	typical
SWE-bench Pro lab	58.6% ✎	▄·····▄·▄▄▄█ beats 6/6	hard · avg 28%	▇ 0.5%	blunt · at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #19 of 393 * · 2.1% of its score · 2 tests

Beats GPT-6 Astra, Claude Sonnet 5.5, GPT-6 Sol here, though they rank higher overall. Trails Gemini 3.6 Flash, Claude Opus 4.6, Muse Spark 1.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1481.4 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 358/379	avg 1268	▇ 1.9%	blunt
IFEval (aggregation unspecified)	95.0% ✎	█·██·█···█·█ beats 4/5	easy · avg 74%	▇ 0.2%	maxed out

math▁▂▃▅▆▅▄█▄▁▃▁ #11 of 398 · 37% of its score · 18 tests

Beats Claude Opus 5, GPT-5.4 Pro here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tier 4 v2	72.5% ✎	▁▂▁▂█▅▂▃▁▁▁▁ beats 53/63	very hard · avg 0.1%	▇▇▇ 8.2%	sharp · at its level
FrontierMath tiers 1–3 v2	85.3% ✎	▂··▁▁▃▄▄█▅▃▃ beats 71/82	very hard · avg 1.2%	▇▇ 5.6%	sharp
MathArena ArXivMath June 2026 (48 problems)	83.6% ✎	▂···▄·██▆▆▄▂ beats 15/19	very hard · avg 4.6%	▇▇ 3.8%	sharp
MathArena ArXivMath April 2026 (40 problems)	68.8% ✎	▁·····▅██▄·▁ beats 17/20	very hard · avg 7.2%	▇ 3.7%	sharp · at its level
MathArena ArXivMath February 2026 (32 problems)	73.4% ✎	▁▁▁▁█▇▄▃▁▃·▃ beats 23/24	very hard · avg 9.1%	▇ 2.3%	at its level
MathArena BrokenArXiv April 2026 (61 problems)	72.1% ✎	▂▂▆█▃▃▂▃▂··▂ beats 16/18	very hard · avg 0.7%	▇ 1.9%	sharp · at its level · noisy
MathArena ArXivMath March 2026 (30 problems)	77.5% ✎	█▃▅·█▃▃·▅·▃▃ beats 14/14	very hard · avg 15%	▇ 1.8%	at its level
MathArena BrokenArXiv May 2026 (50 problems)	50.0% ✎	██·▄▄▄····▄▄ beats 8/8	very hard · avg 1.6%	▇ 1.3%	sharp · at its level

10 more tests

test	result	vs the others in its run	difficulty	counts	why
MathArena BrokenArXiv March 2026 (56 problems)	73.7% ✎	█·▆█▂·▂▂···▃ beats 14/14	very hard · avg 0.8%	▇ 1.3%	sharp · at its level · noisy
MathArena BrokenArXiv June 2026 (54 problems)	69.4% ✎	▄··█▂▂▂▆▄··▂ beats 13/14	very hard · avg 1.1%	▇ 1.3%	sharp · at its level · noisy
MathArena ArXivMath May 2026 (40 problems)	77.5% ✎	▁··▁▄█▄▃·▃·▁ beats 14/17	very hard · avg 10%	▇ 1.3%	at its level
MathArena BrokenArXiv February 2026 (31 problems)	68.1% ✎	▃▂▂▃█▃▂·▃··▃ beats 16/16	very hard · avg 0.7%	▇ 1.3%	sharp · at its level · noisy
MathArena ArXivMath January 2026 (23 problems)	73.9% ✎	▂··▄▂▄█▆▆▄▄█ beats 22/23	hard · avg 22%	▇ 1.3%	typical
MathArena ArXivLean March 2026 (41 problems)	17.1% ✎	▂····▂·····█ beats 4/5	very hard · avg 1.0%	▇ 1.0%	typical
IMO-AnswerBench	79.5% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 21/42	hard · avg 25%	▇ 0.5%	noisy
MathConstraint full release, no tools (329 instances)	66.9% ✎	▄··▄▄·▄▄█▄·▄ beats 8/8	very hard · avg 11%	▇ 0.5%	at its level
FrontierMath: Erdős problems	0.0% ✎	█··········▆ beats 0/6	very hard · avg 0.0%	▇ 0.2%	sharp · too hard for it · noisy
OTIS mock AIME 2024–2025	84.4% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 125/193	very hard · avg 8.5%	▇ 0.1%	sharp · maxed out · noisy

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #21 of 147 · 3.0% of its score · 3 tests

Beats Claude Sonnet 5.5 here, though they rank higher overall. Trails Gemini 3.6 Flash, Gemini 3.5 Flash, Claude Opus 4.6 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena vision (style controlled)	1281.0 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 126/145	avg 1068	▇ 1.3%	blunt
MMMU Pro lab	81.2% ✎	▅·····▃·▃·█▃ beats 6/7	hard · avg 40%	▇ 1.1%	typical
Arena document (style controlled)	1480.2 ✎	▁▁▃▃▄▃▇█▅▇▄▅ beats 30/37	avg 1335	▇ 0.6%	blunt

web search▂▁▂▆▄█▃▃▁▁·▁ #22 of 80 · 6.7% of its score · 6 tests

Trails GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503), GPT-4o (2024-08-06) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp lab	84.4% ✎	▄··█▄··▄·▄█▄ beats 3/8	hard · avg 32%	▇ 1.7%	typical
K-BrowseComp-Verified (Korean web-browsing agent questions, pass@1)	45.7% ✎	▃····█··▃█·▃ beats 8/8	very hard · avg 3.4%	▇ 1.7%	at its level
Humanity's Last Exam with tools lab	52.2% ✎	▃·▃·▅····▅·█ beats 5/8	very hard · avg 12%	▇ 1.2%	at its level
Legal Research Bench (agentic, all-pass)	40.0% ✎	▄█··▄▄▄██·▄█ beats 11/12	very hard · avg 5.8%	▇ 0.8%	at its level
Arena search systems (style controlled)	1223.6 ✎	▁·▁▁▂▂▂█▅▃▁▁ beats 28/30	avg 1055	▇ 0.8%	blunt
RWSearch	85.3% ✎	████··█····█ beats 5/5	medium · avg 44%	▇ 0.6%	typical

other tests20 not in a task · 26% of its score

test	result	vs the others in its run	difficulty	counts	why
EBR-bench (Epoch)	34.3% ✎	▄▂·▆█▂▆▄▆▆·▄ beats 15/23	very hard · avg 0.3%	▇▇ 5.1%	sharp · at its level
Mystery Game Puzzles (Epoch)	56.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 67/74	very hard · avg 9.4%	▇ 3.6%	sharp · at its level · noisy
Furniture Assembly (Epoch)	44.2% ✎	█▆▆▆▆▂▅▂▂▂·▅ beats 20/30	very hard · avg 1.2%	▇ 3.5%	sharp · at its level
Humanity's Last Exam lab	41.4% ✎	▃··▃·▃▃···▃█ beats 5/7	very hard · avg 4.4%	▇ 2.3%	at its level
Chess Puzzles (Epoch)	26.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 99/135	very hard · avg 5.9%	▇ 1.9%	sharp · at its level · noisy
Legal Research Bench (single-shot, no tools, all-pass)	6.8% ✎	▃▃█▃·▅···█▃▃ beats 10/12	very hard · avg 0.8%	▇ 1.2%	too hard for it
EQ-Bench 3 (emotional intelligence, judge Elo)	1327.9 ✎	▁·▂▁▄▅▅▃█▄▁▁ beats 56/62	avg 1044	▇ 1.1%	typical
The Token Games Elo (10-model puzzle-duel tournament)	1167.0 ✎	█··▄█··▄·▄·▄ beats 7/7	avg 815	▇ 1.1%	typical

12 more tests

test	result	vs the others in its run	difficulty	counts	why
WildClawBench	58.2% ✎	▃·▅▅█·▅█▅▃▃▃ beats 16/17	very hard · avg 11%	▇ 1.1%	at its level
GPQA Diamond	90.7% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 177/208	hard · avg 40%	▇ 0.8%	sharp · maxed out
SimpleQA Verified	63.0% ✎	▁▁▃▂▃▃▆▆█▃▃▅ beats 66/78	very hard · avg 8.0%	▇ 0.7%	at its level
FORTE (AGI-Eval)	77.8% ✎	▄···█·····▄█ beats 5/5	medium · avg 37%	▇ 0.7%	typical
WritingBench	84.7% ✎	▁·▁▂▁··▅█▂▂▁ beats 10/20	medium · avg 55%	▇ 0.6%	typical
CyberGym lab	81.8% ✎	▄···▄··▄█▄█▄ beats 6/8	hard · avg 31%	▇ 0.5%	small run
HealthBench Hard lab	33.8% ✎	▂··▂····▂▄·█ beats 6/8	very hard · avg 9.7%	▇ 0.4%	blunt · at its level
MirrorCode (Epoch)	10.0% ✎	█·▄▄·▄·▄·▄·█ beats 1/8	very hard · avg 1.5%	▇ 0.4%	sharp · at its level · noisy
HealthBench lab	58.4% ✎	▃·▃·····█▃▃▅ beats 7/8	hard · avg 24%	▇ 0.4%	blunt · at its level
RobloxQA v2.0 multiple choice (Torpedo Software)	97.5% ✎	▁▁▃▁█▄▃▃▆▃▂▂ beats 43/49	easy · avg 78%	▇ 0.3%	maxed out
NTCIR SciClaimEval Subtask 1 pair-accuracy (dev, single model)	69.9% ✎	▃▃···▃·▃▃█·▃ beats 6/8	medium · avg 45%	▇ 0.3%	blunt · at its level
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	90.7% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 55/98	very easy · avg 90%	▇ 0.0%	blunt

$ where to run it

Through an API

api · $11.25/M · $5 in / $30 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… 30%, MathArena ArXivMath / BrokenA… 22%, OpenAI vendor-reported benchm… 16% and 18 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	▇▇▇▇▇▇▇▇▇ 30%	CC BY 4.0
MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets	12	▇▇▇▇▇▇▇ 22%	CC BY-SA 4.0
OpenAI vendor-reported benchmark scores for its own models (openai.com announcement posts and Deployment Safety Hub system cards) · lab-reported	16	▇▇▇▇▇ 16%	Factual vendor-reported score (not licensed content)
LMArena leaderboard dataset	6	▇▇▇ 12%	CC BY 4.0
meituan-longcat/LongCat-2.0 model card	5	▇ 2%	mit
Agents' Last Exam — authors' own fixed-harness model comparison	1	▇ 2%	CC BY 4.0
Legal Research Bench (Vals AI, arXiv:2610.00609v1)	2	▇ 2%	CC-BY-4.0
K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts (arXiv:2606.02404v1)	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	▇ 1%	CC BY 4.0
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode	2	▇ 1%	MIT
EQ-Bench 3 canonical leaderboard results	1	▇ 1%	MIT
The Token Games: Evaluating Language Model Reasoning with Puzzle Duels — authors' own tournament	1	▇ 1%	CC BY-SA 4.0
WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation (arXiv:2605.10912v1)	1	▇ 1%	CC BY 4.0
zai-org/GLM-5.2 model card	1	▇ 1%	mit
KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy (arXiv:2609.38480v2)	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
MathConstraint: Automated Generation of Verified Combinatorial Reasoning Instances for LLMs — authors' own model runs	1	▇ 0%	CC BY 4.0
TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks — authors' own Terminus-2 model comparison	1	▇ 0%	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
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 80.1