Model profile

GPT-5.4

Name: GPT-5.4
Author: OpenAI

OpenAISupersededReleased Mar 5, 2026

Data verified July 20, 2026

Superseded:OpenAI has released newer models in this line —GPT-5.5

Overall Score

74.24Public #8 of 200Verified #8 of 99

Arena Elo

1466

Eligible category ranks

4of 8

Price (1M tokens)

$2.5 in / $15 out

API pricing

Speed

74tok/s

Context

1.05M

Evidence coverage

52 of 321 tracked benchmarks are published. 23 are verified and 29 provisional. 7 of 8 categories are measured.

Updated July 20, 2026Methodology

Published / tracked: 52 / 321
Verified: 23
Provisional: 29
Categories with evidence: 7 / 8

Agentic17 benchmarks
Mixed evidence
Coding6 benchmarks
Mixed evidence
Reasoning2 benchmarks
Reported
Knowledge13 benchmarks
Mixed evidence
Math2 benchmarks
Verified
Multilingual0 benchmarks
Not measured
Multimodal11 benchmarks
Mixed evidence
Inst. Following1 benchmark
Reported

ProprietaryReasoning

Confidence:

High

base

GPT-5.4 ranks #8 out of 200 models on the public leaderboard with an overall score of 74.24/100. It also ranks #8 out of 99 on the verified leaderboard. This places it in the upper tier of AI models, with competitive scores across most benchmark categories.

GPT-5.4 is a proprietary model with a 1.05M token context window. It uses explicit chain-of-thought reasoning, which typically improves performance on math and complex reasoning tasks at the cost of higher latency and token usage.

GPT-5.4 sits inside the GPT-5.4 family alongside GPT-5.4 mini, GPT-5.4 nano, GPT-5.4 Pro. This profile currently has 52 of 321 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Its strongest category is Knowledge (#10), while its weakest is Coding (#22). This performance profile makes it particularly effective for knowledge-intensive tasks like research, analysis, and factual Q&A.

Peer position

Exact provisional scores and ranks for the closest listed peers. A score can appear before a model clears the evidence threshold for a rank, so equal scores can have different rank states.

Range 71.04–76.72

Grok 4.5
xAI
#776.72
Grok 4.5 is #7 with a score of 76.72.
Compare
GPT-5.4Current model
OpenAI
#874.24
GPT-5.4 is #8 with a score of 74.24.
GPT-5.5
OpenAI
#973.51
GPT-5.5 is #9 with a score of 73.51.
Compare
Qwen3.7 Max
Alibaba
#1072.84
Qwen3.7 Max is #10 with a score of 72.84.
Compare
GPT-5.6 Terra
OpenAI
#1172.57
GPT-5.6 Terra is #11 with a score of 72.57.
Compare
Claude Opus 4.7
Anthropic
#1271.94
Claude Opus 4.7 is #12 with a score of 71.94.
Compare
Muse Spark
Meta
#1371.04
Muse Spark is #13 with a score of 71.04.
Compare

Category percentile

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

Knowledge82%
Eligible cohort rank #10 of 52Category score 82.0
Multimodal54%
Eligible cohort rank #14 of 29Category score 68.9
Agentic85%
Eligible cohort rank #19 of 119Category score 57.0
Coding83%
Eligible cohort rank #22 of 122Category score 59.6

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
Category	Score	Rank	Percentile	Weight	Benchmarks	Evidence
AgenticRank #19 of 119Percentile 85thWeight 22%17 benchmarksMixed sources	57.0	#19 of 119	85th	22%	17 benchmarks	Mixed sources
CodingRank #22 of 122Percentile 83rdWeight 20%6 benchmarksMixed sources	59.6	#22 of 122	83rd	20%	6 benchmarks	Mixed sources
ReasoningRank Not rankedWeight 17%2 benchmarksReported	85.2	Not ranked	Not available	17%	2 benchmarks	Reported
KnowledgeRank #10 of 52Percentile 82ndWeight 12%13 benchmarksMixed sources	82.0	#10 of 52	82nd	12%	13 benchmarks	Mixed sources
MathRank Not rankedWeight 5%2 benchmarksVerified	65.9	Not ranked	Not available	5%	2 benchmarks	Verified
MultilingualWeight 7%0 benchmarksNot measured	Not measured	Not ranked	Not available	7%	0 benchmarks	Not measured
MultimodalRank #14 of 29Percentile 54thWeight 12%11 benchmarksMixed sources	68.9	#14 of 29	54th	12%	11 benchmarks	Mixed sources
Inst. FollowingRank Not rankedWeight 5%1 benchmarkReported	85.2	Not ranked	Not available	5%	1 benchmark	Reported

Chatbot Arena performance

Scroll horizontally to inspect confidence intervals and vote counts.

Chatbot Arena Elo, confidence interval, and vote count by evaluation view
View	Elo	Confidence interval	Votes
Text Overall	1466	±3.9	59,508
Coding	1513	±6.2	16,474
Math	1461	±11.1	3,150
Instruction Following	1462	±5.7	19,968
Creative Writing	1433	±7.3	9,585
Multi-turn	1480	±6.9	11,132
Hard Prompts	1489	±4.8	38,841
Hard Prompts (English)	1488	±5.8	19,082
Longer Query	1479	±5.6	25,154

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Agentic17 benchmarks

Terminal-Bench 2.0Provider exact

75.1%Weighted 38%

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

OSWorld-VerifiedProvider exact

75%Weighted 34%

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

BrowseCompProvider exact

82.7%Weighted 28%

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

CyberGymProvider exact

79.0%Display only

Source: OpenAI: Introducing GPT-5.5Provenance: OpenAI reports GPT-5.4 at 79.0% on CyberGym in the GPT-5.5 launch comparison table.

MCP AtlasProvider exact

70.6%Display only

Source: OpenAI: Introducing GPT-5.5 comparison tableProvenance: Provider exact

ToolathlonProvider exact

54.6%Display only

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

τ²-bench resultsProvider exact

τ²-Bench Tool-Agent-User Evaluation

87.1%Display only

Source: OpenAI: Introducing GPT-5.4 mini and nanoProvenance: OpenAI reports GPT-5.4 at 98.9% on τ²-bench Telecom at xhigh reasoning effort.

Claw-EvalBenchmark exact

60.3%Display only

Source: Claw-Eval leaderboardProvenance: Claw-Eval reports this model as gpt54 in the official 2026-05-09 leaderboard snapshot. BenchLM stores the primary Pass^3 value on the local Claw-Eval display key.

DeepSearchQASecondary exact

73.6%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

AA Agentic IndexReported

Artificial Analysis Agentic Index

41.1%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

APEX-Agents-AAReported

33.3%Display only

Source: Artificial Analysis: apex-agents-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.

GDPval-AAReported

GDPval-AA normalized

44.7%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

GDPval-AAReported

1395Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

Gert LabsBenchmark exact

Gert Labs Composite Game Benchmark

64.89%Display only

Source: Gert Labs rankingsProvenance: Gert Labs reports this composite leaderboard score in the public rankings API. BenchLM scales the source gscore from 0-1 to 0-100 and stores it as a display-only agentic benchmark.

ResearchClawBenchBenchmark exact

15.3%Display only

Source: ResearchClawBench leaderboardProvenance: ResearchClawBench reports this model as ResearchHarness (GPT-5.4) in the official Pass@1 leaderboard. BenchLM stores the one-decimal RADS average on the local ResearchClawBench display key and excludes it from weighted rankings.

JobBenchBenchmark exact

38.9%Display only

Source: JobBench paperProvenance: JobBench reports GPT-5.4 under Codex CLI on the main-set leaderboard. BenchLM stores the reported main-set score.

ExploitGymBenchmark exact

6.0%Display only

Source: ExploitGym paperProvenance: ExploitGym reports GPT-5.4 with Codex CLI produced 54 successful exploits out of 898 tasks. BenchLM stores this as a percentage of the full suite.

Coding6 benchmarks

SWE-bench ProProvider exact

57.7%Weighted 10%

Source: OpenAI: Introducing GPT-5.4Provenance: OpenAI publishes GPT-5.4 exact values for SWE-Bench Pro (Public), Terminal-Bench 2.0, OSWorld-Verified, MMMU Pro, BrowseComp, MCP Atlas, Toolathlon, Tau2-bench Telecom, and HLE without tools in the launch table.

LiveCodeBench ProSecondary exact

87.5%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

React Native EvalsBenchmark exact

85.3%Display only

Source: React Native Evals leaderboardProvenance: React Native Evals reports this exact overall score for GPT 5.4 in the public dashboard run finished on 2026-04-28.

Vibe Code BenchBenchmark exact

Vibe Code Bench v1.1

67.42%Display only

Source: Vals AI: Vibe Code Bench v1.1Provenance: Vals Vibe Code Bench v1.1 reports this exact row under openai/gpt-5.4-2026-03-05; BenchLM stores it on the local vibeCodeBench key.

AA Coding IndexReported

Artificial Analysis Coding Index

71.0%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-SciCodeReported

Artificial Analysis SciCode

56.6%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

Reasoning2 benchmarks

AA-LCRReported

Artificial Analysis Long Context Reasoning

74.0%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

CritPtReported

Critical Physics Tasks

23.4%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

Knowledge13 benchmarks

HLEProvider exact

Humanity's Last Exam

52.1%Weighted 45%

Source: OpenAI: Introducing GPT-5.5 comparison tableProvenance: Provider exact

GPQAReported

Graduate-Level Google-Proof Q&A

92.8%Weighted 7%

Source: Reported upstream sourceProvenance: Reported row carried from an upstream public source. Displayable on BenchLM, but not treated as verified unless explicitly marked otherwise.

HLE w/o toolsProvider exact

Humanity's Last Exam without tools

39.8%Display only

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

GPQA-DSecondary exact

GPQA Diamond

92.8%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

HealthBench HardSecondary exact

40.1%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

MedXpertQA (Text)Secondary exact

MedXpertQA Text

59.6%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

Artificial Analysis Intelligence IndexReported

51.4%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-GPQA DiamondReported

Artificial Analysis GPQA Diamond

92.0%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-HLEReported

Artificial Analysis Humanity's Last Exam

41.6%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-Omniscience IndexReported

Artificial Analysis Omniscience Index

5.7%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-Omniscience AccuracyReported

Artificial Analysis Omniscience Accuracy

50.0%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

AA-Omniscience Hallucination RateReported

Artificial Analysis Omniscience Hallucination Rate

88.6%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

HealthBench ProfessionalBenchmark exact

48.1%Display only

Source: HealthBench Professional paperProvenance: HealthBench Professional reports base GPT-5.4 at 48.1 overall. BenchLM stores the base-model row rather than the ChatGPT for Clinicians harness row.

Math2 benchmarks

FrontierMath v2 (Tiers 1-3)Benchmark exact

FrontierMath v2 Tiers 1-3

47.600%Weighted 30%

Source: Epoch AI FrontierMath v2 leaderboardProvenance: Epoch AI reports FrontierMath v2 Tiers 1-3 at 47.6% for gpt-5.4-2026-03-05_xhigh. BenchLM selects the highest published thinking effort for the model and stores the v2 benchmark slice separately.

FrontierMath v2 (Tier 4)Benchmark exact

FrontierMath v2 Tier 4

27.100%Weighted 10%

Source: Epoch AI FrontierMath v2 leaderboardProvenance: Epoch AI reports FrontierMath v2 Tier 4 at 27.1% for gpt-5.4-2026-03-05_xhigh. BenchLM selects the highest published thinking effort for the model and stores the v2 benchmark slice separately.

Multimodal11 benchmarks

MMMU-ProProvider exact

Massive Multi-discipline Multimodal Understanding Pro

81.2%Weighted 45%

Source: OpenAI: Introducing GPT-5.4Provenance: Provider exact

OfficeQA ProProvider exact

53.2%Weighted 30%

Source: OpenAI: Introducing GPT-5.5 comparison tableProvenance: Provider exact

CharXivSecondary exact

CharXiv Reasoning

82.8%Weighted 25%

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

MMMU-Pro w/ PythonProvider exact

MMMU-Pro with Python

82.1%Display only

Source: OpenAI: Introducing GPT-5.5 comparison tableProvenance: Provider exact

ERQASecondary exact

65.4%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

SimpleVQASecondary exact

61.1%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

ScreenSpot ProSecondary exact

85.4%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

ZeroBenchSecondary exact

41.0%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

MedXpertQA (MM)Secondary exact

MedXpertQA Multimodal

77.1%Display only

Source: Meta AI: Muse Spark comparison chartProvenance: Secondary exact

AA-MMMU-ProReported

Artificial Analysis MMMU-Pro

78.4%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

Design Arena WebsiteReported

Design Arena Website Elo

1252Display only

Source: OpenRouter model benchmarksProvenance: Display-only Design Arena Website Elo synced from OpenRouter model benchmark metadata. It is excluded from BenchLM weighted scoring.

Inst. Following1 benchmark

AA-IFBenchReported

Artificial Analysis IFBench

73.9%Display only

Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.

GPT-5.4 Family

Base entry

GPT-5.4 miniScore 56.77 GPT-5.4 nanoScore 66.79 GPT-5.4 ProScore 60.89

Frequently Asked Questions

How does GPT-5.4 perform overall in AI benchmarks?

GPT-5.4 currently ranks #8 out of 200 models on BenchLM's provisional leaderboard with an overall score of 74.24. It also ranks #8 out of 99 on the verified leaderboard. It is created by OpenAI. Its published context window is 1.05M.

Is GPT-5.4 good for knowledge and understanding?

GPT-5.4 ranks #10 out of 52 models in knowledge and understanding benchmarks with an average score of 82. It is among the top performers in this category.

Is GPT-5.4 good for coding and programming?

GPT-5.4 ranks #22 out of 122 models in coding and programming benchmarks with an average score of 59.6. There are stronger options in this category.

Is GPT-5.4 good for mathematics?

GPT-5.4 has visible benchmark coverage in mathematics, but BenchLM does not currently assign it a global category rank there.

Is GPT-5.4 good for reasoning and logic?

GPT-5.4 has visible benchmark coverage in reasoning and logic, but BenchLM does not currently assign it a global category rank there.

Is GPT-5.4 good for agentic tool use and computer tasks?

GPT-5.4 ranks #19 out of 119 models in agentic tool use and computer tasks benchmarks with an average score of 57. There are stronger options in this category.

Is GPT-5.4 good for multimodal and grounded tasks?

GPT-5.4 ranks #14 out of 29 models in multimodal and grounded tasks benchmarks with an average score of 68.9. There are stronger options in this category.

Is GPT-5.4 good for instruction following?

GPT-5.4 has visible benchmark coverage in instruction following, but BenchLM does not currently assign it a global category rank there.

Which sibling models are related to GPT-5.4?

GPT-5.4 belongs to the GPT-5.4 family. Related variants on BenchLM include GPT-5.4 mini, GPT-5.4 nano, GPT-5.4 Pro.

Does GPT-5.4 have full benchmark coverage on BenchLM?

Not yet. GPT-5.4 currently has 52 published benchmark scores out of the 321 benchmarks BenchLM tracks. BenchLM only exposes non-generated public benchmark rows, so missing categories stay blank until a sourced evaluation is available.

What is the context window size of GPT-5.4?

GPT-5.4 has a published context window of 1.05M, which determines how much text it can process in a single interaction.

Related Resources

Last updated: July 20, 2026 · Runtime metrics stay blank until BenchLM has a sourced snapshot.

Choose with this week’s evidence

Join 2,000+ readers for ranking moves, new releases, pricing changes, and the evidence behind them.

Free. One email per week.

GPT-5.4

Evidence coverage

Evidence by category

Peer position

Category percentile

Category evidence

Chatbot Arena performance

Benchmark Details

GPT-5.4 Family

Frequently Asked Questions

How does GPT-5.4 perform overall in AI benchmarks?

Is GPT-5.4 good for knowledge and understanding?

Is GPT-5.4 good for coding and programming?

Is GPT-5.4 good for mathematics?

Is GPT-5.4 good for reasoning and logic?

Is GPT-5.4 good for agentic tool use and computer tasks?

Is GPT-5.4 good for multimodal and grounded tasks?

Is GPT-5.4 good for instruction following?

Which sibling models are related to GPT-5.4?

Does GPT-5.4 have full benchmark coverage on BenchLM?

What is the context window size of GPT-5.4?

Related Resources

Choose with this week’s evidence