Inference
EN
OpenAI

GPT-5.4

GPT-5.4 is a reasoning model from OpenAI. 67 benchmarks count toward its index, in 7 categories.

  • Closed weights
  • Reasoning
  • Input: text · image · file
  • Released 2026-03
RankedEnough results for an index.
IndexAll categories, out of 100.64.8±2.5Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.38/sMedian 67/s
First tokenSeconds to the first output token.1.7 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$2.50batch $1.25Median $0.60
Output / 1MUS dollars per 1M output tokens.$15.00batch $7.50 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1452±4 #49Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-5.4
Max outputMaximum output tokens per request.
128K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.25
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
7 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
DatabricksCloud$2.50$15.001.05M
DigitalOceanCloud$2.50$15.00400K
Microsoft AzureCloud$2.50$15.001.05M
OpenRouterRouter$2.50$15.001.05M
SAP AI CoreCloud$2.50$15.001.05M
Snowflake CortexCloud$2.50$15.001.05M
Amazon BedrockCloud$2.75$16.501M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.21 counted62.159 / 218
CodingCode writing and repair.11 counted61.860 / 209
ReasoningLogic problems and puzzles.10 counted69.023 / 227
MultimodalTasks with images and text.8 counted62.138 / 84
KnowledgeFacts and expert knowledge.8 counted66.436 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted63.410 / 63
MathMath problems.7 counted68.024 / 168

Results

67 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
τ²-Bench Tool-Agent-User EvaluationAgentic98.9%2 / 149
AIMEMath96.7% ±0.54 / 95
FrontierMath-2025-02-28-PrivateMath47.6% ±2.94 / 69
Terminal-Bench 2.0 · BenchLMAgentic75.1%3 / 41
Gert Labs Composite Game BenchmarkAgentic64.9%4 / 52
GPQA diamond · Epoch AIKnowledge93.3% ±1.819 / 217
Graduate-Level Google-Proof Q&AKnowledge92.8%13 / 141
IOI v1Coding67.8% ±9.96 / 61
Critical Physics TasksReasoningAbove fitFar above this model's index.23.4%23 / 214
Artificial Analysis Omniscience AccuracyKnowledge50.8%18 / 164
Artificial Analysis GPQA DiamondKnowledge92.0%22 / 199
FrontierMath-Tier-4-2025-07-01-PrivateMath27.1% ±6.47 / 55
GPQA Diamond · Vals AIKnowledge91.7% ±1.918 / 135
Artificial Analysis Long Context ReasoningReasoning82.0%29 / 217
Artificial Analysis Humanity's Last ExamKnowledge43.7%31 / 222
ScreenSpot ProMultimodal85.4%3 / 20
Terminal-Bench 2.0 · Vals AIAgentic58.4% ±5.310 / 66
MMMU ProMultimodal87.5% ±0.814 / 92
HealthBench HardKnowledge40.1%2 / 13
GPQA Diamond · BenchLMKnowledge92.8%14 / 84
Artificial Analysis IFBenchInstruction73.9%26 / 146
ARC-AGI-2 (semi-private)Reasoning74.0%24 / 131
React Native EvalsCoding85.3%3 / 16
LiveBench Data AnalysisReasoning79.3%15 / 80
Massive Multi-discipline Multimodal Understanding ProMultimodal81.2%10 / 53
MMLU ProKnowledge87.5% ±0.426 / 135
ZeroBenchMultimodal41.0%2 / 10
ARC-AGI-1 (semi-private)Reasoning93.7%26 / 126
Chess PuzzlesReasoning48.0% ±5.019 / 90
Terminal-Bench HardAgentic57.6%5 / 23
OTIS Mock AIME 2024-2025Math95.3% ±3.246 / 200
FrontierMath-Tiers-1-3-v2-PrivateMath78.6% ±2.422 / 90
SWE-Bench verified · Epoch AICoding76.9% ±1.98 / 32
Mystery Game PuzzlesReasoning37.0% ±4.920 / 80
LiveBench MathematicsMath94.1%17 / 66
Humanity's Last ExamKnowledge52.1%20 / 77
ToolathlonAgentic54.6%6 / 22
LiveCodeBenchCoding84.1% ±1.040 / 142
APEX-Agents-AAAgentic33.3%9 / 30
LiveBench ReasoningReasoning88.1%25 / 80
FrontierMath-Tier-4-v2-PrivateMath49.0% ±8.023 / 72
Artificial Analysis MMMU-ProMultimodal78.4%37 / 113
SWE-benchCoding78.2% ±1.830 / 87
LiveBench LanguageKnowledge82.6%23 / 66
GDPval-AA normalizedAgentic37.3%47 / 133
τ²-bench BankingAgentic39.4%11 / 28
LiveBench Instruction FollowingInstruction70.2%27 / 66
BrowseCompAgentic82.7%27 / 63
CharXiv ReasoningMultimodal82.8%22 / 49
OSWorld-VerifiedAgentic75.0%17 / 37
Code MigrationCoding35.0% ±4.136 / 77
SimpleQA VerifiedKnowledge45.1% ±1.643 / 91
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
SWE-bench ProCoding57.7%41 / 83
Agent Arena command recoveryAgentic1.226 / 50
JobBenchAgentic38.9%19 / 36
LiveBench CodingCoding77.5%35 / 66
LiveBench Agentic CodingAgentic53.8%35 / 66
SkillsBenchCoding51.7% ±4.719 / 35
Humanity's Last Exam without toolsKnowledge39.8%29 / 53
Vibe Code Bench v1.1Coding48.5% ±5.169 / 123
EBR-benchReasoning25.4%18 / 31
Furniture AssemblyReasoning37.5% ±6.222 / 37
MCP AtlasAgentic70.6%31 / 52
ERQAMultimodal65.4%11 / 18
CyberGymAgentic79.0%18 / 29
Claw-EvalAgentic60.3%27 / 42
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal61.7%26 / 39
SimpleVQAMultimodal61.1%12 / 18
Artificial Analysis Tau3-BankingAgentic39.6%15 / 22
ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable jobAgentic11.0%14 / 18
ARC-AGI-3 (semi-private)Reasoning0.2%21 / 26
DeepSearchQAAgentic73.6%20 / 24
MirrorCodeCoding23.3% ±8.133 / 39
OfficeQA ProMultimodal53.2%14 / 16
ExploitGymAgentic6.0%21 / 23
HealthBench ProfessionalKnowledge48.1%18 / 18
PostTrainBench v1.1Coding19.0%15 / 15
MedQAHealthcare96.1% ±0.25 / 95
FrontierMath-Tier-4-2025-07-01-PublicMath50.0% ±50.02 / 36
FrontierMath-2025-02-28-PublicMath80.0% ±13.36 / 64
LegalBenchLegal86.0% ±0.417 / 149
ARC-AGI-2 (public eval)Reasoning84.2%20 / 122
MortgageTaxFinance68.3% ±0.917 / 97
CorpFin v2Finance65.3% ±0.928 / 131
ARC-AGI-1 (public eval)Reasoning96.4%25 / 116
Finance Agent v1.1Finance57.2% ±2.911 / 51
ProgramBenchCoding0.5% ±0.519 / 65
TaxEval v2Finance74.0% ±0.942 / 143
CaseLaw v2Legal63.8% ±2.216 / 53
MedCodeHealthcare41.3% ±2.155 / 106
ProgramBenchCoding0.0%34 / 65
MedScribeHealthcare77.5% ±3.369 / 109
LiveCodeBench ProCoding87.5%—
MMMU-Pro with PythonMultimodal82.1%—
MedXpertQA MultimodalMultimodal77.1%—
LifeVal: everyday life tasksAgentic73.0%—
MedXpertQA TextKnowledge59.6%—
BioBasicsBench: everyday wet-lab scienceReasoning33.2%—
ResearchClawBenchAgentic15.3%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYLiveBenchNo licence statedARC Prize FoundationTerms restrict reuseSierra ResearchMITLMArenaCC BY 4.0models.devMIT

History

5 releases
Index65.7 → 64.8
23 Oct 202612 Nov 2026
Output / 1M$15.00 → $15.00
23 Oct 202612 Nov 2026

Questions

How much does GPT-5.4 cost?
$2.50 per 1M input tokens, $15.00 per 1M output tokens.
How good is GPT-5.4?
Its index is 64.8 out of 100. That is place 46 of 283 ranked models.
What is GPT-5.4 best at?
Reasoning: 69.0 out of 100, place 23 of 227.
How fast is GPT-5.4?
38 output tokens per second, 1.7 s to the first token.
What is the context window of GPT-5.4?
1.05M tokens.
Is GPT-5.4 open source?
No. Its weights are not published.
When was GPT-5.4 released?
March 2026.

Badge

GPT-5.4 on Inference360: place 46 of 283, index 64.8
HTML
<a href="https://inference360.ai/models/gpt-5-4"><img src="https://inference360.ai/badges/gpt-5-4.svg" alt="GPT-5.4 on Inference360: place 46 of 283, index 64.8" height="20"></a>
Markdown
[![GPT-5.4 on Inference360: place 46 of 283, index 64.8](https://inference360.ai/badges/gpt-5-4.svg)](https://inference360.ai/models/gpt-5-4)

Compare with

Muse Spark 1.264.9Muse Spark 1.164.6Qwen3.8-Flash-Next64.3dots3-note Preview63.4GPT-5.4 Pro73.3

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from OpenAI

All models
GPT-5.4 Pro73.3GPT-5.3 InstantUnrankedGPT Realtime 1.5UnrankedGPT-5.4 mini54.5GPT-5.4 nano50.9