BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases9

GPT-5.4 vs o3

Full profiles: GPT-5.4 (OpenAI) and o3 (OpenAI): every benchmark, price and rank.

Updated · Data from 18 sources

As of 29 Oct 2026, GPT-5.4 and o3 are neck and neck on quality (63.3 vs 61.0 on the BenchLeader Index, inside the ±4.2 margin), and o3 is 1.6× cheaper at $3.50 vs $5.63 per 1M tokens blended. Choose GPT-5.4 for instruction following (72 vs 65), reasoning (66 vs 61), and multimodal (61 vs 57); choose o3 for fast first answers (5.9 s vs 136.5 s) and agents & tools (70 vs 64) and maths (71 vs 65), where it also streams faster.

Key facts (as of 29 Oct 2026)

  • GPT-5.4 scores 63.3 on the BenchLeader Index and o3 scores 61.0 (BenchLeader, 29 Oct 2026).
  • GPT-5.4 costs $5.63 per 1M tokens blended (list price, 29 Oct 2026).
  • o3 costs $3.50 per 1M tokens blended (list price, 29 Oct 2026).
  • o3 streams faster than GPT-5.4 (Artificial Analysis, 7-day median, 29 Oct 2026).
  • Of the 10 benchmarks both models have, GPT-5.4 leads 10 (BenchLeader, 29 Oct 2026).
GPT-5.4 benchmarks xhigh OpenAI o3 benchmarks 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 62 category scoreMedium
Agents & tool useo370 vs 64 category score; Terminal-Bench Hard 37.1% vs 57.6%Medium
Mathso371 vs 65 category scoreMedium
ReasoningGPT-5.466 vs 61 category score; GPQA Diamond (AA) 92.0% vs 82.7%Medium
KnowledgeGPT-5.459 vs 56 category scoreMedium
Long documentsGPT-5.466 vs 62 category score; AA-LCR v1.1 82.0% vs 74.7%Medium
Images and other mediaGPT-5.461 vs 57 category score; MMMU-Pro 78.4% vs 70.1%Medium
Following instructionsGPT-5.472 vs 65 category score; IFBench 74.0% vs 71.4%Medium
Chat quality (human preference)–Only o3 has an Arena rating (61 category score)Not comparable
Lowest costo3$3.50 vs $5.63 per 1M blended; agentic coding session $0.152 vs $0.210–
Fastest first answero35.9 s (o3) vs 136.5 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 o3 are neck and neck on quality (63.3 vs 61.0, within the ±4.2 margin).
  • GPT-5.4 (xhigh) is stronger in coding, instruction following, knowledge, long context, multimodal, reasoning.
  • o3 is stronger in agents & tools, maths, human preference.
  • o3 is 1.6× cheaper ($3.50 vs $5.63 per 1M blended).
  • o3 streams 1.7× faster (158 vs 94 tokens per second).

At a glance

SpecGPT-5.4o3
DeveloperOpenAIOpenAI
Released5 Mar 202616 Apr 2025
Context window (tokens)1,050,000200,000
Compared atGPT-5.4 (xhigh)o3
Open weightsNoNo
Price in / out / cached input per 1M$2.50 / $15.00 / $0.250$2.00 / $8.00 / $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)o3Leader
Reasoning66.160.9GPT-5.4
Coding64.561.8GPT-5.4
Agents & tools63.670.3o3
Maths65.170.5o3
Knowledge58.656.1GPT-5.4
Instruction following72.365.1GPT-5.4
Human preference–61.1not comparable
Multimodal61.156.7GPT-5.4
Long context65.962.1GPT-5.4

Where sources disagree

Across the 10 benchmarks both have, GPT-5.4 leads 10, o3 leads 0. Every overall index that rates both puts GPT-5.4 ahead.

Overall indexGPT-5.4 (xhigh)o3Leader
BenchLeader Index63.361.0neck and neck
Arena Text1475†1432GPT-5.4

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)o3
Blended price per 1M tokens$5.63$3.50
Input / output per 1M$2.50 / $15.00$2.00 / $8.00
Cached input per 1M$0.250$0.500
Price per task: chat reply$0.0055$0.0032
Price per task: summarize a 30-page report$0.039$0.029
Price per task: code edit$0.037$0.024
Price per task: agentic coding session$0.210$0.152
Price per task: structured extraction$0.0080$0.0056
Output speed94 tok/s158 tok/s
Time to first answer136.5 s5.9 s
Time to a complete answer141.9 s9.1 s

Each model is compared at its best-ranked reasoning-effort setting. GPT-5.4 (xhigh) scores 63.3 and takes 137 s to its first answer token; GPT-5.4 (low) scores 58.9 in 2 s, GPT-5.4 (high) scores 58.8 in 1 s. o3 scores 61.0 and takes 6 s to its first answer token; o3 (high) scores 53.8 in 6 s, o3 (medium) scores 53.3 in 6 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

140 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)o3
BenchLeader IndexBenchLeader63.361.0
Agents & tools scoreBenchLeader63.670.3
Coding scoreBenchLeader64.561.8
Instruction following scoreBenchLeader72.365.1
Knowledge scoreBenchLeader58.656.1
Long context scoreBenchLeader65.962.1
Maths scoreBenchLeader65.170.5
Multimodal scoreBenchLeader61.156.7
Reasoning scoreBenchLeader66.160.9
Human preference scoreBenchLeader–61.1
Blended price $/MBenchLeader$5.63$3.50
Output speedBenchLeader94 tok/s158 tok/s
Time to first answerBenchLeader136.5 s5.9 s
Context windowBenchLeader1,050,000 tokens200,000 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub93.3%81.8%†
Humanity's Last ExamScale AI / CAIS36.2%20.3%†
SimpleBenchSimpleBench–53.1%†
Arena Hard PromptsArena (LMArena)1487†1441
LiveBench 1.0 ReasoningLiveBench88.1%–
GPQA Diamond (AA)Artificial Analysis92.0%82.7%
Humanity's Last Exam (AA)Artificial Analysis43.7%20.1%
GPQA Diamond (Vals)Vals AI91.7%84.1%†
Kagi LLM BenchmarkKagi LLM Benchmark63.8%†67.6%
GPQA Diamond (HELM)HELM Capabilities–75.3%
ARC-AGI-1ARC Prize93.7%60.8%†
ARC-AGI-2ARC Prize74.0%6.5%†
ARC-AGI-3ARC Prize0.2%†–
CritPtArtificial Analysis23.4%1.1%
SciPredictScale AI SEAL–17.9%
Chess PuzzlesEpoch AI Benchmarking Hub48.0%36.0%†
Mystery Game PuzzlesEpoch AI Benchmarking Hub37.0%29.0%†
CL-benchEpoch AI Benchmarking Hub27.9%17.8%†
CL-bench LifeEpoch AI Benchmarking Hub21.7%–
LMCAEpoch AI Benchmarking Hub52.0%39.7%†
DTBenchEpoch AI Benchmarking Hub94.4%84.8%†
ForecastBenchForecastBench63.1%†64.6%
Coding
SWE-bench Verified (Epoch)Epoch AI Benchmarking Hub76.9%†62.3%†
SciCodeSciCode56.6%–
WeirdMLWeirdML77.7%52.4%†
Show all 140 benchmarks
MetricPublisherGPT-5.4 (xhigh)o3
Coding
GSO-BenchGSO-Bench31.4%8.8%†
Arena CodingArena (LMArena)1511†1460
Arena WebDevArena (LMArena)1401†–
LiveBench 1.0 CodingLiveBench77.5%–
LiveCodeBenchVals AI84.1%83.9%†
SWE-bench (Vals)Vals AI78.2%–
SWE-Bench ProScale AI SEAL59.1%–
Aider PolyglotAider polyglot leaderboard–81.3%
SWE-bench Verified (bash only)SWE-bench–58.4%
SWE-bench Verified (any scaffold)SWE-bench–58.4%
Code MigrationVals AI35.0%–
ProgramBenchVals AI0.0%–
Vibe Code Bench v1.1Vals AI67.4%–
SWE-Bench Pro (private)Scale AI SEAL59.1%–
SWE Atlas: Codebase QnAScale AI SEAL40.8%–
SWE Atlas: RefactoringScale AI SEAL44.3%–
SWE Atlas: Test WritingScale AI SEAL44.4%–
MirrorCodeEpoch AI Benchmarking Hub15.6%†–
CadEvalEpoch AI Benchmarking Hub–74.0%†
DeepSWE v1.1Epoch AI Benchmarking Hub51.8%–
ALE-BenchEpoch AI Benchmarking Hub1607†933.5†
AlgoTuneAlgoTune1.9†–
Terminal-Bench 2.1 (AA)Artificial Analysis78.3%–
LiveCodeBench (AA)Artificial Analysis–80.8%
Agents & tools
Terminal-BenchTerminal-Bench81.8%†–
GDPvalOpenAI–30.8%†
APEX-AgentsMercor52.4%†–
LiveBench Agentic CodingLiveBench53.8%–
Terminal-Bench HardArtificial Analysis57.6%37.1%
τ²-Bench Telecom (AA)Artificial Analysis87.1%80.7%
MCP AtlasScale AI SEAL70.6%–
HiL-BenchScale AI SEAL9.7%†–
BFCL OverallBerkeley Function Calling Leaderboard–63.0%
τ³-bench Bankingτ²-bench39.4%–
τ³-Banking (AA)Artificial Analysis39.6%–
APEX-Agents (AA)Artificial Analysis33.3%–
Harvey's Legal Agent Benchmark v1Vals AI0.0%–
SkillsBenchVals AI51.7%–
Terminal-Bench 2.0 (Vals)Vals AI58.4%†–
DeepResearch BenchDeepResearch Bench35.1%†45.2%†
PostTrainBenchEpoch AI Benchmarking Hub19.0%†–
METR Time HorizonsMETR74.3%63.6%
GBAEvalEpoch AI Benchmarking Hub45.1%†–
Vending-Bench 2Andon Labs6144.2†–
BrowseComp-PlusBrowseComp-Plus–50.5%
DeepSearchQAGoogle63.7%†–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub78.6%33.3%†
FrontierMath Tier 4Epoch AI Benchmarking Hub49.0%–
OTIS Mock AIMEEpoch AI Benchmarking Hub95.3%83.9%†
MATH Level 5Epoch AI Benchmarking Hub–97.8%†
ProofBenchVals AI56.0%–
LiveBench MathematicsLiveBench94.2%–
AIME (Vals)Vals AI96.7%85.3%†
MGSMVals AI–91.8%†
AIME 2026MathArena99.2%–
HMMT February 2026MathArena97.7%–
USAMO 2026MathArena95.2%–
IMO 2025MathArena–16.7%†
MathArena ApexMathArena54.2%–
Omni-MATH (HELM)HELM Capabilities–71.4%
MATH 500Vals AI–94.6%†
Arena MathsArena (LMArena)1460†1448
AIME 2025 (AA)Artificial Analysis–88.3%
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub45.1%49.4%†
LiveBench 1.0 Data AnalysisLiveBench79.3%–
MMLU-ProVals AI87.5%85.6%†
LegalBenchVals AI86.0%83.8%†
CorpFinVals AI65.3%59.7%†
TaxEvalVals AI74.0%74.6%†
MedQAVals AI96.1%96.1%†
PRBench FinanceScale AI SEAL45.6%†47.7%
PRBench LegalScale AI SEAL44.4%†48.6%
MultiNRCScale AI SEAL58.3%†45.5%†
MMLU-Pro (HELM)HELM Capabilities–85.9%
CaseLaw v2Vals AI63.8%–
MedCodeVals AI41.3%47.3%†
MedScribeVals AI77.5%76.7%†
EBR-benchEpoch AI Benchmarking Hub25.4%–
Instruction following
LiveBench LanguageLiveBench82.6%–
LiveBench Instruction FollowingLiveBench70.2%–
IFBenchArtificial Analysis74.0%71.4%
MultiChallengeScale AI SEAL–56.6%†
TutorBenchScale AI SEAL–52.1%†
IFEval (HELM)HELM Capabilities–86.9%
WildBench (HELM)HELM Capabilities–86.1%
Arena Instruction FollowingArena (LMArena)1461†1403
Human preference
Arena TextArena (LMArena)1465†1432
EQ-Bench 4EQ-Bench1258†–
Arena Creative WritingArena (LMArena)1434†1383
Arena Multi-turnArena (LMArena)1480†1420
Arena Longer QueriesArena (LMArena)1477†1411
Lech Mazur Creative WritingLech Mazur–83.9%†
Multimodal
Arena VisionArena (LMArena)1292†1214
MMMU-ProArtificial Analysis78.4%70.1%
VISTAScale AI SEAL50.9%†50.1%†
MMMU (validation)MMMU–82.9%
MMMU-Pro (official)MMMU–76.4%
MMMU-Pro (Vals)Vals AI87.5%80.4%†
MortgageTaxVals AI68.3%65.7%†
VTBScale AI SEAL29.2%†13.7%
GeoBenchGeoBench–60.0%†
VPCTEpoch AI Benchmarking Hub–52.0%†
Blueprint-Bench 2Epoch AI Benchmarking Hub27.1%†–
Long context
Fiction.LiveBench 120kFiction.live–100.0%†
AA-LCR v1.1Artificial Analysis82.0%74.7%
Arena DocumentArena (LMArena)1470†–
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub157.2†146.8
LiveBench 1.0LiveBench78.0%–
AA Intelligence Index v4.3.2Artificial Analysis39.020.2
AA Progress Index v1Artificial Analysis311.6258.3
HELM Capabilities meanHELM Capabilities–81.1%
Vals RSI IndexVals AI15.5%–
Safety & honesty
FORTRESSScale AI SEAL–16.0%
MASKScale AI SEAL89.7%†84.5%†
PropensityBenchScale AI SEAL–10.5%

Data as of 2026-10-29. Best configuration of each model; every score links to its source on the model pages.

What changed

  • o3: 2 new results (LiveCodeBench (AA), AIME 2025 (AA)).
  • GPT-5.4 (xhigh): 1 new result (AA Progress Index v1).
  • o3: 1 new result (AA Progress Index v1).
  • GPT-5.4 (xhigh): 2 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index).
  • o3: 1 score revision.
  • GPT-5.4 (xhigh): 2 new results (USAMO 2026, τ³-bench Banking).
  • GPT-5.4 (xhigh): 1 score revision.
  • o3: 1 new result (PRBench Legal).
  • GPT-5.4 (xhigh): 1 new result (SWE Atlas: Codebase QnA).
  • o3: 1 new result (VTB).

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

GPT-5.4 vs o3: questions

Is GPT-5.4 better than o3?
GPT-5.4 (xhigh) and o3 are neck and neck on quality (63.3 vs 61.0, within the ±4.2 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 29 Oct 2026, so check the category scores for your use.
GPT-5.4 vs o3: which is cheaper, and will that change?
o3 is cheaper today: $3.50 against $5.63 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Is GPT-5.4 or o3 better for coding?
GPT-5.4 scores higher in coding (65 vs 62 on the category index, where 50 is average).
Which is faster, GPT-5.4 or o3?
o3 streams faster (158 vs 94 output tokens per second, Artificial Analysis 7-day median).
Is GPT-5.4 or o3 better for agents and tool use?
o3 scores higher in agentic tasks (70 vs 64 on the category index, where 50 is average). On Terminal-Bench Hard, the top shared agents & tools benchmark, it is 57.6% for GPT-5.4 against 37.1% for o3.
More questions (4)
Which is better for long documents?
GPT-5.4 scores higher on long-context benchmarks (66 vs 62). GPT-5.4 accepts more context: 1,050,000 against 200,000 tokens for o3.
Which has lower latency?
o3 answers first: 5.9 s against 136.5 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.
Do independent benchmarks agree on GPT-5.4 vs o3?
Across the 10 benchmarks both have, GPT-5.4 leads 10, o3 leads 0. Every overall index that rates both puts GPT-5.4 ahead.
When was GPT-5.4 released?
GPT-5.4 was released on 5 Mar 2026.