OpenAIReasoning model

GPT-5.5

Reasoning effort

GPT-5.5 is an OpenAI proprietary reasoning model, released 23 Apr 2026. Its best configuration (xhigh reasoning effort) ranks #37 of 756 on the BenchLeader Index at 65.9 ±3.6, in the upper half. The ± is the point: 98 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in reasoning (76) and lowest in agents & tools (61). At $11.25 per million tokens blended it is among the most expensive ranked models. Output speed of 85 tokens per second puts it faster than most, with a first answer in 52.3 s. It has been measured at 6 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 2 Oct 2026.

Blended price: $11.25/M; $5.00 in · $30.00 out
Output speed: 85 tok/s; measured by Artificial Analysis
First answer: 52 s; first token 3.10 s
Context: 1.1M
Full answer: 58 s; median, reasoning included
Cost per run: $2.63; one full Intelligence Index run
Released: 23 Apr 2026

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

Overall index66
Reasoning76
Coding62
Agents & tools61
Maths70
Knowledge65
Instruction following74
Multimodal63
Long context68
Composite69

Versions

OpenAI has shipped 12 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
GPT-5.6	9 Jul 2026	–	–
GPT-5.5xhighthis page	23 Apr 2026	65.9	#37
GPT-5.4xhigh	5 Mar 2026	64.2	#54
GPT-5.2xhigh	11 Dec 2025	60.9	#97
GPT-5.1high	12 Nov 2025	58.0	#155
GPT-5high	7 Aug 2025	58.1	#151
GPT-4.1	14 Apr 2025	48.6	#404
GPT-4.5	27 Feb 2025	49.9	#362
GPT 4 0125	25 Jan 2024	45.7	#485
GPT-4	6 Nov 2023	44.3	#518
GPT 4 0613	13 Jun 2023	39.9	#645
GPT 4 0314	14 Mar 2023	43.5	#543

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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	53.1	#280	81 tok/s	0.92 s	$0.011	Agents & tools 54 · Coding 56 · Composite 56 · Instruction following 48 · Knowledge 60 · Long context 57 · Maths 49 · Multimodal 54 · Reasoning 50
low	60.0	#111	84 tok/s	1.59 s	$0.011	Agents & tools 62 · Coding 57 · Composite 65 · Instruction following 64 · Knowledge 69 · Long context 66 · Maths 61 · Multimodal 62 · Reasoning 56
medium	63.1	#70	83 tok/s	7.03 s	$0.011	Agents & tools 67 · Coding 60 · Composite 68 · Instruction following 70 · Knowledge 70 · Long context 67 · Multimodal 64 · Reasoning 67
high	65.7	#38	87 tok/s	21 s	$0.011	Agents & tools 70 · Coding 62 · Composite 72 · Human preference 68 · Instruction following 70 · Knowledge 70 · Long context 68 · Maths 68 · Multimodal 64 · Reasoning 69
xhighbest	65.9	#37	85 tok/s	52 s	$0.011	Agents & tools 61 · Coding 62 · Composite 69 · Instruction following 74 · Knowledge 65 · Long context 68 · Maths 70 · Multimodal 63 · Reasoning 76
not stated	62.3	#85	60 tok/s	3.10 s	$0.011	Agents & tools 62 · Coding 55 · Human preference 67 · Maths 69 · Multimodal 65 · Reasoning 70

What more thinking costs

Turning the effort up buys index points and multiplies the bill. Here xhigh costs 2.9× medium for +2.8 on the index — 1.8 points per doubling of spend. Compare that with other models.

How its index has moved

8 Sept 2026 to 5 Oct 2026

xhigh
high
medium
not stated

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

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	not stated	Source
GPQA Diamond	77.3%#130	90.7%#38	–	–	–	–	Epoch AI Benchmarking Hub
SimpleBench	–	–	–	–	–	69.0%#15	SimpleBench
LMArena Hard Prompts	–	–	–	1501#31	–	1499#35	LMArena
LiveBench Reasoningnot in index	–	–	–	–	89.7%#14	–	LiveBench
GPQA Diamond (AA)not in index	76.8%#222	91.0%#54	92.6%#32	93.2%#22	93.5%#14	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	13.7%#248	32.7%#125	42.4%#61	45.0%#47	45.8%#45	–	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	–	–	93.2%#10	–	Vals AI
Kagi LLM Benchmark	–	–	–	–	–	88.8%#2	Kagi LLM Benchmark
ARC-AGI-1	–	76.2%#110	92.2%#52	94.5%#35	95.0%#33	–	ARC Prize
ARC-AGI-2	–	33.3%#107	70.4%#52	83.3%#34	85.0%#28	–	ARC Prize
ARC-AGI-3	–	–	–	0.4%#34	–	–	ARC Prize
CritPt	1.4%#190	8.0%#121	18.6%#61	25.4%#40	27.1%#31	–	Artificial Analysis
Chess Puzzlesnot in index	10.0%#133	26.0%#61	–	–	–	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	18.0%#64	28.0%#34	–	52.0%#10	56.0%#8	–	Epoch AI Benchmarking Hub
CL-bench Lifenot in index	–	–	–	22.2%#1	–	–	Epoch AI Benchmarking Hub
LMCAnot in index	–	–	–	–	54.3%#33	–	Epoch AI Benchmarking Hub
DTBenchnot in index	–	–	–	–	96.0%#22	–	Epoch AI Benchmarking Hub
ForecastBenchnot in index	–	–	–	–	–	60.6%#25	ForecastBench
BTF-3not in index	–	–	–	14.3%#3	–	–	Epoch AI Benchmarking Hub

Coding

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
SciCode	47.3%#114	51.6%#82	53.5%#69	55.9%#39	56.1%#37	–	SciCode
WeirdML	67.2%#38	–	–	83.9%#16	84.9%#14	–	WeirdML
FrontierCode	–	–	–	–	–	43.0%#16	Cognition
GSO-Bench	–	–	–	–	40.2%#9	–	GSO-Bench
LMArena Coding	–	–	–	1519#40	–	1511#60	LMArena
LMArena WebDev	–	–	–	1487#55	1512#49	1456#63	LMArena
LiveBench Codingnot in index	–	–	–	–	82.2%#10	–	LiveBench
SciCode (AA)not in index	–	–	54.5%#56	56.1%#38	55.8%#40	–	Artificial Analysis
LiveCodeBench	–	–	–	–	85.3%#33	–	Vals AI
SWE-bench (Vals)not in index	–	–	–	–	82.6%#19	–	Vals AI
Code Migrationnot in index	–	–	–	–	45.2%#15	–	Vals AI
ProgramBenchnot in index	–	–	–	–	0.5%#13	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	–	–	69.8%#33	–	Vals AI
SWE Atlas: Codebase QnAnot in index	–	–	–	–	45.4%#8	–	Scale AI SEAL
SWE Atlas: Refactoringnot in index	–	–	–	–	44.8%#6	–	Scale AI SEAL
SWE Atlas: Test Writingnot in index	–	–	–	–	42.6%#9	–	Scale AI SEAL
MirrorCodenot in index	–	–	–	10.0%#8	–	–	Epoch AI Benchmarking Hub
Surface Evolver Benchnot in index	–	–	81.3%#8	88.1%#5	–	–	Epoch AI Benchmarking Hub
DeepSWEnot in index	–	27.0%#65	54.0%#40	64.4%#29	67.0%#22	–	Epoch AI Benchmarking Hub
ALE-Benchnot in index	1127.6#43	–	1589.4#14	–	1943.0#9	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	–	–	5.0%#88	9.1%#78	14.7%#60	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	61.0%#92	65.5%#79	80.5%#42	79.4%#48	84.3%#30	–	Artificial Analysis

Agents & tools

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
Terminal-Bench	–	–	–	–	–	84.7%#1	Terminal-Bench
OSWorld-Verified 2.0	–	–	–	–	13.0%#9	–	OSWorld
Remote Labor Index	–	–	–	–	–	6.3%#5	Scale AI / CAIS
APEX-Agents	–	–	–	–	–	55.1%#19	Mercor
LMArena Agent	–	–	–	–	4.2#14	1#28	LMArena
LiveBench Agentic Codingnot in index	–	–	–	–	54.0%#31	–	LiveBench
Terminal-Bench Hard	49.2%#24	52.3%#19	57.6%#11	59.9%#9	60.6%#7	–	Artificial Analysis
τ²-Bench Telecom (AA)not in index	69.3%#157	83.9%#104	91.8%#56	93.0%#47	93.9%#39	–	Artificial Analysis
Terminal-Bench 2.1 (Vals)	–	–	–	76.4%#16	–	–	Vals AI
MCP Atlas	–	–	–	–	75.3%#18	–	Scale AI SEAL
HiL-Bench	–	–	–	–	–	39.7%#7	Scale AI SEAL
GDPval (AA)not in index	22.9%#150	26.6%#135	35.9%#104	41.5%#82	41.8%#81	–	Artificial Analysis
τ²-Bench Banking (AA)	14.8%#117	24.9%#81	29.9%#64	36.7%#42	39.0%#34	–	Artificial Analysis
ITBench SRE (AA)not in index	–	–	–	–	45.8%#13	–	Artificial Analysis
Analyst Agent (AA)not in index	7.5%#35	–	–	–	50.0%#7	–	Artificial Analysis
APEX-Agents (AA)not in index	–	–	–	–	37.7%#6	–	Artificial Analysis
Finance Agent v2not in index	–	–	–	–	51.8%#32	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	–	–	–	–	3.8%#33	–	Vals AI
Legal Research Benchnot in index	–	–	–	–	40.4%#24	–	Vals AI
SkillsBenchnot in index	–	–	–	–	62.2%#5	–	Vals AI
SREBenchnot in index	–	–	–	–	3.8%#11	–	Vals AI
Tax Agent Benchnot in index	–	–	–	–	60.5%#30	–	Vals AI
Terminal-Bench 2.0 (Vals)not in index	–	–	–	73.2%#1	–	–	Vals AI
Time Horizon Index: KSPnot in index	–	–	–	–	8.3%#8	–	Vals AI
DeepResearch Benchnot in index	–	48.7%#17	49.6%#12	54.0%#4	–	–	DeepResearch Bench
PostTrainBenchnot in index	–	–	–	–	27.2%#9	–	Epoch AI Benchmarking Hub
ExploitBenchnot in index	–	–	–	–	–	47.4%#2	Epoch AI Benchmarking Hub
GBAEvalnot in index	–	–	–	–	–	53.2%#6	Epoch AI Benchmarking Hub
Vending-Bench 2not in index	–	–	–	–	–	7523.8#13	Andon Labs

Maths

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
FrontierMath Tiers 1–3	–	–	–	–	85.3%#12	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	–	–	72.5%#18	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	57.8%#159	84.4%#90	–	–	–	–	Epoch AI Benchmarking Hub
ProofBench	–	–	–	–	50.0%#28	–	Vals AI
LiveBench Mathematicsnot in index	–	–	–	–	95.9%#13	–	LiveBench
AIME 2026	–	–	–	–	100.0%#1	–	MathArena
HMMT February 2026	–	–	–	–	98.5%#1	–	MathArena
MathArena Apex	–	–	–	–	80.2%#2	–	MathArena
LMArena Maths	–	–	–	1493#23	–	1500#17	LMArena

Knowledge

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
SimpleQA Verified	–	–	–	–	63.0%#13	–	Epoch AI Benchmarking Hub
LiveBench Data Analysisnot in index	–	–	–	–	81.6%#4	–	LiveBench
AA-Omniscience	-4.8#148	15.1#77	18.1#71	18.8#70	20.5#63	–	Artificial Analysis
MMLU-Pro	–	–	–	–	88.1%#21	–	Vals AI
LegalBench	–	–	–	–	86.5%#12	–	Vals AI
CorpFin	–	–	–	–	68.4%#8	–	Vals AI
TaxEval	–	–	–	–	75.0%#24	–	Vals AI
CaseLaw v2not in index	–	–	–	–	66.2%#7	–	Vals AI
Excel Modeling Benchmarknot in index	–	–	–	–	64.5%#23	–	Vals AI
MedCodenot in index	–	–	–	–	49.1%#24	–	Vals AI
MedScribenot in index	–	–	–	–	86.9%#18	–	Vals AI
Public Benefits Benchnot in index	–	–	–	–	60.9%#27	–	Vals AI
DrugDiscoveryBenchnot in index	–	–	–	–	51.6%#5	–	Scale AI SEAL
EBR-benchnot in index	–	–	–	–	34.3%#9	–	Epoch AI Benchmarking Hub
GDP.pdfnot in index	–	–	–	–	26.0%#12	–	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	45.6%#85	54.9%#37	56.8%#33	57.0%#32	58.0%#28	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	7.3%#461	12.0%#364	10.4%#390	10.9%#382	11.0%#380	–	Artificial Analysis

Instruction following

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
LiveBench Languagenot in index	–	–	–	–	87.4%#9	–	LiveBench
LiveBench Instruction Followingnot in index	–	–	–	–	70.7%#26	–	LiveBench
IFBench	46.1%#194	64.3%#108	71.0%#63	71.6%#52	75.8%#26	–	Artificial Analysis
LMArena Instruction Followingnot in index	–	–	–	1478#23	–	1475#26	LMArena

Human preference

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
LMArena Text	–	–	–	1481#25	–	1477#30	LMArena
EQ-Bench 4	–	–	–	–	–	1315#4	EQ-Bench
LMArena Creative Writingnot in index	–	–	–	1451#35	–	1448#41	LMArena
LMArena Multi-turnnot in index	–	–	–	1487#26	–	1484#31	LMArena
LMArena Longer Queriesnot in index	–	–	–	1487#29	–	1487#33	LMArena

Multimodal

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
LMArena Vision	–	–	–	1292#24	–	1297#19	LMArena
MMMU-Pro	71.4%#131	79.0%#61	81.2%#39	81.1%#40	79.9%#52	–	Artificial Analysis
MMMU-Pro (Vals)not in index	–	–	–	–	88.3%#8	–	Vals AI
MortgageTaxnot in index	–	–	–	–	68.8%#10	–	Vals AI
SAGEnot in index	–	–	–	–	51.5%#13	–	Vals AI
Blueprint-Bench 2not in index	–	–	–	–	–	36.2%#8	Epoch AI Benchmarking Hub

Long context

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source	Trend
AA-LCR	64.0%#241	81.0%#54	83.0%#24	84.3%#8	84.3%#8	–	Artificial Analysis
LMArena Documentnot in index	–	–	–	1480#11	–	1483#10	LMArena

Composite

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source
Epoch Capabilities Indexnot in index	–	–	–	–	–	159.2#10	Epoch AI Benchmarking Hub
LiveBench	–	–	–	–	80.2%#11	–	LiveBench
AA Intelligence Index	23.2#174	30.7#104	33.8#88	37.0#73	38.4#67	–	Artificial Analysis
Vals Multimodal Indexnot in index	–	–	–	–	68.1%#9	–	Vals AI

Safety & honesty

Benchmark	no reasoning	low	medium	high	xhigh	not stated	Source	Trend
FORTRESSnot in index	–	–	–	–	16.3%#49	–	Scale AI SEAL

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

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
Amazon Bedrock	81 tok/s	2.06 s	$5.50	$33.00	1.1M	–
OpenAI	67 tok/s	3.24 s	$5.00	$30.00	1.1M	–
Azure	54 tok/s	5.32 s	$5.00	$30.00	1.1M	–

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 $0.500 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.011	$0.0097	55.8 s
Summarise a 30-page report	12,000 / 600	$0.078	$0.037	59.3 s
Code edit	6,000 / 1,500	$0.075	$0.055	1.2 min
Agentic coding session	60,000 / 4,000	$0.420	$0.217	1.7 min
Structured extraction	2,000 / 200	$0.016	$0.0093	54.6 s