OpenAIReasoning model

GPT-5.2

Name: GPT-5.2 benchmark results
Creator: BenchLeader

Best configuration ranks #65 of 610 on the BenchLeader Index at 61.5 ±3.6. Last measured 8 Sept 2026. Released 11 Dec 2025.

Blended price: $4.81/M; $1.75 in · $14.00 out
Output speed: 72 tok/s
First answer: 138 s; first token 2.31 s
Context: 400k

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index62
Reasoning61
Coding53
Agents & tools61
Knowledge61
Instruction following67
Human preference68
Multimodal60
Long context68
Composite67

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	50.1	#292	65 tok/s	0.97 s	$0.0049	Agents & tools 40 · Coding 50 · Composite 50 · Instruction following 49 · Knowledge 58 · Long context 49 · Maths 51 · Multimodal 50 · Reasoning 54
low	50.2	#290	72 tok/s	138 s	$0.0049	Coding 50 · Knowledge 44 · Maths 59 · Reasoning 49
medium	58.5	#123	72 tok/s	138 s	$0.0049	Agents & tools 62 · Coding 59 · Composite 62 · Instruction following 64 · Knowledge 54 · Long context 62 · Maths 65 · Multimodal 59 · Reasoning 55
high	55.5	#175	72 tok/s	138 s	$0.0049	Agents & tools 55 · Coding 61 · Composite 50 · Human preference 63 · Knowledge 45 · Maths 62 · Multimodal 59 · Reasoning 56
xhigh	57.3	#141	72 tok/s	138 s	$0.0049	Agents & tools 51 · Coding 65 · Knowledge 56 · Maths 58 · Reasoning 62
defaultbest	61.5	#65	72 tok/s	138 s	$0.0049	Agents & tools 61 · Coding 53 · Composite 67 · Human preference 68 · Instruction following 67 · Knowledge 61 · Long context 68 · Multimodal 60 · Reasoning 61

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
GPQA Diamond	73.2%#145	82.7%#99	87.9%#53	88.2%#52	91.4%#26	–	Epoch AI Benchmarking Hub
Humanity's Last Exam	–	–	–	–	–	27.8%#10	Scale AI / CAIS
SimpleBench	–	–	–	45.8%#52	–	45.8%#52	SimpleBench
LMArena Hard Prompts	–	–	–	1460#80	–	1497#30	LMArena
LiveBench Reasoningnot in index	–	–	–	83.2%#33	–	–	LiveBench
GPQA Diamond (AA)not in index	71.2%#278	–	86.4%#111	–	–	90.3%#63	Artificial Analysis
Humanity's Last Exam (AA)not in index	8.0%#282	–	26.7%#129	–	–	37.7%#67	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	–	–	91.7%#18	–	Vals AI
Kagi LLM Benchmark	–	–	–	–	–	73.3%#18	Kagi LLM Benchmark
ARC-AGI-1	–	55.7%#117	72.7%#96	78.7%#84	86.2%#68	94.5%#24	ARC Prize
ARC-AGI-2	–	9.7%#107	26.7%#92	43.3%#77	52.9%#69	72.9%#33	ARC Prize

Coding

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
SWE-bench Verified (Epoch)	–	–	–	73.8%#17	–	–	Epoch AI Benchmarking Hub
WeirdML	49.6%#67	49.6%#67	63.4%#39	–	72.2%#27	–	WeirdML
GSO-Bench	–	–	–	27.4%#9	–	–	GSO-Bench
LMArena Coding	–	–	–	1490#78	–	1515#35	LMArena
LMArena WebDev	–	–	–	–	–	1417#65	LMArena
LiveBench Codingnot in index	–	–	–	76.1%#33	–	–	LiveBench
LiveCodeBench	–	–	–	–	85.4%#31	–	Vals AI
IOI	–	–	–	–	54.8%#8	–	Vals AI
SWE-bench (Vals)not in index	–	–	–	–	75.8%#41	–	Vals AI
SWE-Bench Pro	–	–	–	–	–	29.9%#16	Scale AI SEAL
SWE-bench Verified (bash only)	–	–	–	72.8%#7	–	69.0%#14	SWE-bench
SWE-bench Verified (any scaffold)not in index	–	–	–	72.8%#11	–	–	SWE-bench

Agents & tools

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
Terminal-Bench	–	–	64.9%#9	–	–	64.9%#9	Terminal-Bench
GDPval	49.7%#1	–	–	–	–	–	OpenAI
Remote Labor Index	–	–	2.5%#7	–	–	2.1%#8	Scale AI / CAIS
APEX-Agents	–	–	–	23.0%#33	34.4%#19	23.0%#33	Mercor
LiveBench Agentic Codingnot in index	–	–	–	50.3%#31	–	–	LiveBench
Terminal-Bench Hard	31.8%#99	–	43.2%#39	–	–	47.0%#27	Artificial Analysis
τ²-Bench Telecom (AA)not in index	46.5%#199	–	74.3%#136	–	–	84.8%#95	Artificial Analysis
MCP Atlas	–	–	–	–	67.6%#21	–	Scale AI SEAL
BFCL Overall	–	–	–	–	–	55.9%#14	Berkeley Function Calling Leaderboard
τ²-bench	61.6%#8	–	–	84.8%#3	–	–	τ²-bench

Maths

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
FrontierMath Tiers 1–3	–	–	–	–	67.4%#20	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	–	–	31.7%#27	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	62.2%#143	78.9%#105	93.9%#42	96.1%#25	96.1%#27	–	Epoch AI Benchmarking Hub
ProofBench	–	–	–	–	15.0%#44	–	Vals AI
LiveBench Mathematicsnot in index	–	–	–	93.2%#13	–	–	LiveBench
AIME (Vals)	–	–	–	–	96.9%#2	–	Vals AI
MGSM	–	–	–	–	94.0%#6	–	Vals AI
AIME 2026	–	–	–	98.3%#4	–	–	MathArena
HMMT February 2026	–	–	–	97.0%#3	–	–	MathArena
MathArena Apex	–	–	–	13.5%#18	–	–	MathArena

Knowledge

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
SimpleQA Verified	–	32.8%#51	32.7%#53	34.3%#43	37.1%#39	–	Epoch AI Benchmarking Hub
LiveBench Data Analysisnot in index	–	–	–	78.2%#19	–	–	LiveBench
AA-Omniscience	-12.3#154	–	0.3#92	–	–	-0.9#102	Artificial Analysis
MMLU-Pro	–	–	–	–	86.2%#44	–	Vals AI
LegalBench	–	–	–	–	82.8%#58	–	Vals AI
CorpFin	–	–	–	–	65.9%#25	–	Vals AI
TaxEval	–	–	–	–	75.8%#11	–	Vals AI
MedQA	–	–	–	–	94.1%#19	–	Vals AI
MultiNRC	–	–	–	–	–	42.2%#17	Scale AI SEAL

Instruction following

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
LiveBench Languagenot in index	–	–	–	79.8%#23	–	–	LiveBench
IFBench	47.4%#185	–	65.2%#102	–	–	75.4%#31	Artificial Analysis
TutorBench	–	–	–	–	–	53.5%#11	Scale AI SEAL

Human preference

Benchmark	no reasoning	low	medium	high	xhigh	default	Source	Trend
LMArena Text	–	–	–	1438#83	–	1476#26	LMArena

Multimodal

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
LMArena Vision	–	–	–	1247#52	–	1268#35	LMArena
MMMU-Pro	65.8%#140	–	74.6%#82	–	–	–	Artificial Analysis
VISTA	–	–	–	–	–	46.6%#20	Scale AI SEAL

Long context

Benchmark	no reasoning	low	medium	high	xhigh	default	Source	Trend
AA-LCR	45.7%#266	–	70.3%#157	–	–	82.7%#20	Artificial Analysis

Composite

Benchmark	no reasoning	low	medium	high	xhigh	default	Source
Epoch Capabilities Indexnot in index	–	–	–	–	–	153.5#32	Epoch AI Benchmarking Hub
LiveBench	–	–	–	74.6%#30	–	–	LiveBench
AA Intelligence Index	17.0#202	–	26.5#105	–	–	30.4#77	Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
OpenAI Fast	69 tok/s	2.31 s	$3.50	$28.00	400k	–
OpenAI	43 tok/s	1.77 s	$1.75	$14.00	400k	–
Azure	34 tok/s	4.61 s	$1.75	$14.00	400k	–

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.175 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0049	$0.0044	2.4 min
Summarise a 30-page report	12,000 / 600	$0.029	$0.015	2.4 min
Code edit	6,000 / 1,500	$0.032	$0.024	2.6 min
Agentic coding session	60,000 / 4,000	$0.161	$0.090	3.2 min
Structured extraction	2,000 / 200	$0.0063	$0.0039	2.3 min