BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

GPT-5.4 vs GPT-5.5

Full profiles: GPT-5.4 (OpenAI) and GPT-5.5 (OpenAI): every benchmark, price and rank.

Updated · Data from 11 sources

As of 25 Oct 2026, GPT-5.4 (xhigh) and GPT-5.5 (xhigh) are neck and neck on quality (64.7 vs 66.1 on the BenchLeader Index, inside the ±4.5 margin of error), and GPT-5.4 (xhigh) costs 2.0× less, which makes it the better buy for most workloads. Pick GPT-5.4 for coding (65 vs 61); pick GPT-5.5 for reasoning (76 vs 69).

GPT-5.4 benchmarks xhigh OpenAI GPT-5.5 benchmarks xhigh OpenAIOpen in the comparison tool

Which to pick for…

The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.

Use casePickWhyConfidence
CodingGPT-5.465 vs 61 category score; SciCode 56.6% vs 55.8%Medium
Agents & tool useGPT-5.464 vs 62 category score; τ³-Banking (AA) 39.6% vs 39.0%Medium
MathsGPT-5.570 vs 65 category score; FrontierMath Tiers 1–3 85.3% vs 78.6%Medium
ReasoningGPT-5.576 vs 69 category score; ARC-AGI-2 85.0% vs 74.0%Medium
KnowledgeGPT-5.565 vs 60 category score; SimpleQA Verified 63.0% vs 45.1%Medium
Long documentsGPT-5.567 vs 66 category score; AA-LCR 84.3% vs 82.0%Medium
Images and other mediaGPT-5.563 vs 61 category score; MMMU-Pro 79.9% vs 78.4%Medium
Following instructionsGPT-5.574 vs 72 category score; IFBench 75.8% vs 74.0%Medium
Lowest costGPT-5.4$5.63 vs $11.25 per 1M blended; agentic coding session $0.210 vs $0.420–
Fastest first answerGPT-5.539.1 s (GPT-5.5 (xhigh)) vs 118.0 s (GPT-5.4 (xhigh)) to the first answer token–
High volume on a budgetNeither flagshipA smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs–

Key differences

  • GPT-5.4 (xhigh) and GPT-5.5 (xhigh) are neck and neck on quality (64.7 vs 66.1, within the ±4.5 margin).
  • GPT-5.4 (xhigh) is stronger in agents & tools, coding, composite.
  • GPT-5.5 (xhigh) is stronger in instruction following, knowledge, long context, maths, multimodal, reasoning.
  • GPT-5.4 (xhigh) is 2.0× cheaper ($5.63 vs $11.25 per 1M blended).

At a glance

SpecGPT-5.4GPT-5.5
DeveloperOpenAIOpenAI
Released5 Mar 202623 Apr 2026
Context window (tokens)1,050,0001,050,000
Compared atGPT-5.4 (xhigh)GPT-5.5 (xhigh)
Open weightsNoNo
Price in / out / cached input per 1M$2.50 / $15.00 / $0.250$5.00 / $30.00 / $0.500

Category by category

Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.

CategoryGPT-5.4 (xhigh)GPT-5.5 (xhigh)Leader
Reasoning69.175.5GPT-5.5
Coding64.661.1GPT-5.4
Agents & tools63.662.3GPT-5.4
Maths65.269.7GPT-5.5
Knowledge59.765.3GPT-5.5
Instruction following72.373.9GPT-5.5
Multimodal61.462.9GPT-5.5
Long context66.167.3GPT-5.5
Composite73.172.4level

Where sources disagree

Across the 65 benchmarks both have, GPT-5.4 leads 4, GPT-5.5 leads 61. Independent sources split: GPT-5.4 leads on AA Intelligence Index v4.3.2; GPT-5.5 leads on Arena Text.

Overall indexGPT-5.4 (xhigh)GPT-5.5 (xhigh)Leader
BenchLeader Index64.766.1neck and neck
AA Intelligence Index v4.3.239.038.4GPT-5.4
Arena Text1475†1482GPT-5.5

Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).

Price, speed and latency

List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.

MeasureGPT-5.4 (xhigh)GPT-5.5 (xhigh)
Blended price per 1M tokens$5.63$11.25
Input / output per 1M$2.50 / $15.00$5.00 / $30.00
Cached input per 1M$0.250$0.500
Price per task: chat reply$0.0055$0.011
Price per task: summarize a 30-page report$0.039$0.078
Price per task: code edit$0.037$0.075
Price per task: agentic coding session$0.210$0.420
Price per task: structured extraction$0.0080$0.016
Cost to run a full benchmark suite–$2.63
Output speed96 tok/s86 tok/s
Time to first answer118.0 s39.1 s
Time to a complete answer123.2 s45.0 s

Each model is compared at its best-ranked reasoning-effort setting. GPT-5.4 (xhigh) scores 64.7 and takes 118 s to its first answer token; GPT-5.4 (low) scores 59.7 in 2 s, GPT-5.4 (high) scores 58.9 in 1 s. GPT-5.5 (xhigh) scores 66.1 and takes 39 s to its first answer token; GPT-5.5 (high) scores 65.3 in 14 s, GPT-5.5 (medium) scores 62.9 in 3 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.

Every benchmark

148 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.

MetricPublisherGPT-5.4 (xhigh)GPT-5.5 (xhigh)
BenchLeader IndexBenchLeader64.766.1
Agents & tools scoreBenchLeader63.662.3
Coding scoreBenchLeader64.661.1
Composite scoreBenchLeader73.172.4
Instruction following scoreBenchLeader72.373.9
Knowledge scoreBenchLeader59.765.3
Long context scoreBenchLeader66.167.3
Maths scoreBenchLeader65.269.7
Multimodal scoreBenchLeader61.462.9
Reasoning scoreBenchLeader69.175.5
Blended price $/MBenchLeader$5.63$11.25
Output speedBenchLeader96 tok/s86 tok/s
Time to first answerBenchLeader118.0 s39.1 s
Context windowBenchLeader1,050,000 tokens1,050,000 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub93.3%90.7%†
Humanity's Last ExamScale AI / CAIS36.2%–
SimpleBenchSimpleBench–69.0%†
Arena Hard PromptsArena (LMArena)1487†1499†
LiveBench 2.0 ReasoningLiveBench–89.7%†
LiveBench 1.0 ReasoningLiveBench88.1%89.7%
GPQA Diamond (AA)Artificial Analysis92.0%93.5%
Humanity's Last Exam (AA)Artificial Analysis43.7%45.8%
GPQA Diamond (Vals)Vals AI91.7%93.2%
Kagi LLM BenchmarkKagi LLM Benchmark63.8%†88.8%†
ARC-AGI-1ARC Prize93.7%95.0%
ARC-AGI-2ARC Prize74.0%85.0%
ARC-AGI-3ARC Prize0.2%†0.4%†
CritPtArtificial Analysis23.4%27.1%
Chess PuzzlesEpoch AI Benchmarking Hub48.0%63.0%
Mystery Game PuzzlesEpoch AI Benchmarking Hub37.0%56.0%
CL-benchEpoch AI Benchmarking Hub27.9%–
CL-bench LifeEpoch AI Benchmarking Hub21.7%22.2%†
LMCAEpoch AI Benchmarking Hub52.0%54.3%
DTBenchEpoch AI Benchmarking Hub94.4%96.0%
ForecastBenchForecastBench63.1%†64.4%†
BTF-3Epoch AI Benchmarking Hub–14.3%†
Coding
SWE-bench Verified (Epoch)Epoch AI Benchmarking Hub76.9%†–
SciCodeSciCode56.6%55.8%
WeirdMLWeirdML77.7%84.9%
Show all 148 benchmarks
MetricPublisherGPT-5.4 (xhigh)GPT-5.5 (xhigh)
Coding
FrontierCodeCognition–43.0%†
GSO-BenchGSO-Bench31.4%40.2%
Arena CodingArena (LMArena)1511†1511†
Arena WebDevArena (LMArena)1401†1513
LiveBench 2.0 CodingLiveBench–85.1%†
LiveBench 1.0 CodingLiveBench77.5%82.2%
SciCode (AA)Artificial Analysis–55.8%
LiveCodeBenchVals AI84.1%85.3%
SWE-bench (Vals)Vals AI78.2%82.6%
SWE-Bench ProScale AI SEAL59.1%–
Code MigrationVals AI35.0%45.2%
ProgramBenchVals AI0.0%0.5%
Vibe Code Bench v1.1Vals AI67.4%69.8%
SWE-Bench Pro (private)Scale AI SEAL59.1%–
SWE Atlas: Codebase QnAScale AI SEAL40.8%45.4%
SWE Atlas: RefactoringScale AI SEAL44.3%44.8%
SWE Atlas: Test WritingScale AI SEAL44.4%42.6%
MirrorCodeEpoch AI Benchmarking Hub15.6%†10.0%†
Surface Evolver BenchEpoch AI Benchmarking Hub–88.1%†
DeepSWE v1.1Epoch AI Benchmarking Hub51.8%67.0%
ALE-BenchEpoch AI Benchmarking Hub1607†1943.0
AlgoTuneAlgoTune1.9†–
Terminal-Bench 4.0 (AA)Artificial Analysis–14.7%
Terminal-Bench 2.1 (AA)Artificial Analysis78.3%84.3%
Agents & tools
Terminal-BenchTerminal-Bench81.8%†84.7%†
OSWorld-Verified 2.0OSWorld–13.0%
Remote Labor IndexScale AI / CAIS–6.3%†
APEX-AgentsMercor52.4%†55.1%†
LiveBench Long AgenticLiveBench–44.6%†
LiveBench BrowserLiveBench–48.0%†
LiveBench Agentic CodingLiveBench53.8%54.0%
Terminal-Bench HardArtificial Analysis57.6%60.6%
τ²-Bench Telecom (AA)Artificial Analysis87.1%93.9%
Terminal-Bench 2.1 (Vals)Vals AI–76.4%†
MCP AtlasScale AI SEAL70.6%75.3%
HiL-BenchScale AI SEAL9.7%†39.7%†
τ³-bench Bankingτ²-bench39.4%44.6%
τ³-Banking (AA)Artificial Analysis39.6%39.0%
ITBench SRE (AA)Artificial Analysis–45.8%
Analyst Agent (AA)Artificial Analysis–50.0%
APEX-Agents (AA)Artificial Analysis33.3%37.7%
Finance Agent v2Vals AI–51.8%
Harvey's Legal Agent Benchmark v1Vals AI0.0%3.8%
Legal Research BenchVals AI–40.4%
SkillsBenchVals AI51.7%62.2%
SREBenchVals AI–3.8%
Tax Agent BenchVals AI–21.8%
Terminal-Bench 2.0 (Vals)Vals AI58.4%†73.2%†
Time Horizon Index: KSPVals AI–8.3%
DeepResearch BenchDeepResearch Bench35.1%†54.0%†
PostTrainBenchEpoch AI Benchmarking Hub19.0%†27.2%
ExploitBenchEpoch AI Benchmarking Hub–47.4%†
METR Time HorizonsMETR74.3%–
GBAEvalEpoch AI Benchmarking Hub45.1%†53.2%†
Vending-Bench 2Andon Labs6144.2†7523.8†
AutomationBenchArtificial Analysis–47.3%
GDP.pdf (AA)Artificial Analysis–21.2%
EnterpriseOps-GymArtificial Analysis–46.6%
AA-Briefcase v1.1Artificial Analysis–1136
DeepSearchQAGoogle63.7%†–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub78.6%85.3%
FrontierMath Tier 4Epoch AI Benchmarking Hub49.0%72.5%
OTIS Mock AIMEEpoch AI Benchmarking Hub95.3%84.4%†
ProofBenchVals AI56.0%50.0%
LiveBench MathematicsLiveBench94.2%95.9%
AIME (Vals)Vals AI96.7%–
AIME 2026MathArena99.2%100.0%
HMMT February 2026MathArena97.7%98.5%
USAMO 2026MathArena95.2%98.2%
MathArena ApexMathArena54.2%80.2%
Arena MathsArena (LMArena)1460†1501†
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub45.1%63.0%
LiveBench 2.0 Data AnalysisLiveBench–81.6%†
LiveBench ResearchLiveBench–61.6%†
LiveBench 1.0 Data AnalysisLiveBench79.3%81.6%
AA-OmniscienceArtificial Analysis5.820.5
MMLU-ProVals AI87.5%88.1%
LegalBenchVals AI86.0%86.5%
CorpFinVals AI65.3%68.4%
TaxEvalVals AI74.0%75.0%
MedQAVals AI96.1%–
PRBench FinanceScale AI SEAL45.6%†–
PRBench LegalScale AI SEAL44.4%†–
MultiNRCScale AI SEAL58.3%†–
CaseLaw v2Vals AI63.8%66.2%
Excel Modeling BenchmarkVals AI–64.5%
MedCodeVals AI41.3%49.1%
MedScribeVals AI77.5%86.9%
Public Benefits Bench v1.1Vals AI–60.9%
DrugDiscoveryBenchScale AI SEAL–51.6%
EBR-benchEpoch AI Benchmarking Hub25.4%34.3%
GDP.pdfEpoch AI Benchmarking Hub–26.0%
AA-Omniscience: accuracyArtificial Analysis50.9%58.0%
AA-Omniscience: non-hallucinationArtificial Analysis8.3%11.0%
Instruction following
LiveBench LanguageLiveBench82.6%87.4%
LiveBench Instruction FollowingLiveBench70.2%70.7%
IFBenchArtificial Analysis74.0%75.8%
Arena Instruction FollowingArena (LMArena)1460†1475†
Human preference
Arena TextArena (LMArena)1465†1477†
EQ-Bench 4EQ-Bench1260†1294†
Arena Creative WritingArena (LMArena)1435†1448†
Arena Multi-turnArena (LMArena)1480†1484†
Arena Longer QueriesArena (LMArena)1477†1487†
Multimodal
Arena VisionArena (LMArena)1292†1297†
LiveBench DesignLiveBench–70.3%†
MMMU-ProArtificial Analysis78.4%79.9%
VISTAScale AI SEAL50.9%†–
MMMU-Pro (Vals)Vals AI87.5%88.3%
MortgageTaxVals AI68.3%68.8%
SAGEVals AI43.3%51.5%
VTBScale AI SEAL29.2%†–
Blueprint-Bench 2Epoch AI Benchmarking Hub27.1%†36.2%†
Long context
AA-LCRArtificial Analysis82.0%84.3%
Arena DocumentArena (LMArena)1470†1483†
MLCRArtificial Analysis–28.9%
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub157.3†159.7†
LiveBench 2.0LiveBench–68.7%†
LiveBench 1.0LiveBench78.0%80.2%
AA Intelligence Index v4.3.2Artificial Analysis39.038.4
Vals RSI IndexVals AI15.5%16.5%
Vals Multimodal IndexVals AI–68.1%
Safety & honesty
FORTRESSScale AI SEAL–16.3%
MASKScale AI SEAL89.7%†–

Data as of 2026-10-25. Best configuration of each model; every score links to its source on the model pages.

What changed

  • GPT-5.5 (xhigh): 1 new result (AA-Briefcase v1.1).
  • GPT-5.4 (xhigh): 2 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index).
  • GPT-5.5 (xhigh): 2 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index).
  • GPT-5.5 (xhigh): 1 score revision.
  • GPT-5.4 (xhigh): 2 new results (USAMO 2026, τ³-bench Banking).
  • GPT-5.5 (xhigh): 6 new results (USAMO 2026, AutomationBench, GDP.pdf (AA)…).
  • GPT-5.5 (xhigh): BenchLeader Index 66.2 → 65.7.
  • GPT-5.5 (xhigh): 1 score revision.
  • GPT-5.4 (xhigh): 1 score revision.
  • GPT-5.5 (xhigh): 1 new result (Chess Puzzles); 1 score revision.

Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs

GPT-5.4 vs GPT-5.5: questions

Is GPT-5.4 better than GPT-5.5?
GPT-5.4 (xhigh) and GPT-5.5 (xhigh) are neck and neck on quality (64.7 vs 66.1, within the ±4.5 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-25, so check the category scores for your use.
Which is cheaper, GPT-5.4 or GPT-5.5, and will that change?
GPT-5.4 is cheaper today: $5.63 against $11.25 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Is GPT-5.4 better than GPT-5.5 for coding?
GPT-5.4 scores higher in coding (65 vs 61 on the category index, where 50 is average). On LiveCodeBench, the top shared coding benchmark, it is 84.1% for GPT-5.4 against 85.3% for GPT-5.5.
Is GPT-5.4 better than GPT-5.5 for agents and tool use?
GPT-5.4 scores higher in agentic tasks (64 vs 62 on the category index, where 50 is average). On MCP Atlas, the top shared agents & tools benchmark, it is 70.6% for GPT-5.4 against 75.3% for GPT-5.5.
Which is better for long documents?
GPT-5.5 scores higher on long-context benchmarks (67 vs 66). Both accept 1,050,000 tokens of context.
Which has lower latency?
GPT-5.5 (xhigh) answers first: 39.1 s against 118.0 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
Which is faster, GPT-5.4 or GPT-5.5?
GPT-5.4 streams faster: 96 against 86 output tokens per second.
Do independent benchmarks agree on GPT-5.4 vs GPT-5.5?
Across the 65 benchmarks both have, GPT-5.4 leads 4, GPT-5.5 leads 61. Independent sources split: GPT-5.4 leads on AA Intelligence Index v4.3.2; GPT-5.5 leads on Arena Text.
When was GPT-5.5 released?
GPT-5.5 was released on 23 Apr 2026.