Inference
EN
OpenAI

GPT-5.4 mini

GPT-5.4 mini is a reasoning model from OpenAI in the GPT-5.4 family. 45 benchmarks count toward its index, in 8 categories.

  • Closed weights
  • Reasoning
  • Input: file · image · text
  • Released 2026-03
RankedEnough results for an index.
IndexAll categories, out of 100.54.5±2.4Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.95/sMedian 67/s
First tokenSeconds to the first output token.1.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.75batch $0.375Median $0.60
Output / 1MUS dollars per 1M output tokens.$4.50batch $2.25 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.400KMedian 262K
EloLMArena rating and rank.1412±4 #140Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-5.4-mini
Max outputMaximum output tokens per request.
128K
Knowledge cutoffLast date of training data.
31 Aug 2025
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.075
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
4 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
DatabricksCloud$0.75$4.50400K
DigitalOceanCloud$0.75$4.50400K
Microsoft AzureCloud$0.75$4.50400K
OpenRouterRouter$0.75$4.50400K

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.13 counted56.195 / 218
CodingCode writing and repair.8 counted55.196 / 209
ReasoningLogic problems and puzzles.7 counted52.2100 / 227
MultimodalTasks with images and text.1 counted54.8—
KnowledgeFacts and expert knowledge.6 counted53.7106 / 268
MultilingualTasks in many languages.1 counted57.6—
InstructionTasks with strict prompt rules.2 counted52.632 / 63
MathMath problems.7 counted53.970 / 168

Results

45 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
EuroEval SpanishMultilingual62.8% ±0.49 / 176
EuroEval SwedishMultilingual68.9% ±0.612 / 195
EuroEval PolishMultilingual62.5% ±0.413 / 209
EuroEval GermanMultilingual57.8% ±0.512 / 191
EuroEval SwedishMultilingual67.3% ±0.614 / 195
EuroEval SpanishMultilingual60.8% ±0.414 / 176
EuroEval ItalianMultilingual66.2% ±0.514 / 174
AIMEMath95.6% ±0.48 / 95
EuroEval GermanMultilingual56.3% ±0.517 / 191
EuroEval FrenchMultilingual67.4% ±0.416 / 178
EuroEval DutchMultilingual63.1% ±0.416 / 175
EuroEval FrenchMultilingual67.3% ±0.417 / 178
EuroEval DutchMultilingual63.0% ±0.517 / 175
EuroEval PolishMultilingual60.2% ±0.421 / 209
EuroEval FrenchMultilingual67.1% ±0.418 / 178
EuroEval SwedishMultilingual65.0% ±0.720 / 195
EuroEval ItalianMultilingual65.8% ±0.418 / 174
EuroEval PortugueseMultilingual64.7% ±0.419 / 182
EuroEval SpanishMultilingual57.9% ±0.519 / 176
EuroEval PortugueseMultilingual63.7% ±0.521 / 182
EuroEval PolishMultilingual58.8% ±0.525 / 209
EuroEval DutchMultilingual62.1% ±0.421 / 175
EuroEval GermanMultilingual54.6% ±0.623 / 191
EuroEval ItalianMultilingual62.5% ±0.523 / 174
EuroEval PortugueseMultilingual60.6% ±0.429 / 182
Artificial Analysis IFBenchInstruction73.3%29 / 146
τ²-Bench Tool-Agent-User EvaluationAgentic93.4%31 / 149
Graduate-Level Google-Proof Q&AKnowledge88.0%38 / 141
Artificial Analysis GPQA DiamondKnowledge87.5%56 / 199
FrontierMath-2025-02-28-PrivateMath28.3% ±2.120 / 69
GPQA diamond · Epoch AIKnowledge86.9% ±2.465 / 217
Critical Physics TasksReasoning10.0%65 / 214
OTIS Mock AIME 2024-2025Math88.9% ±4.763 / 200
Terminal-Bench HardAgentic52.3%8 / 23
Terminal-Bench 2.0 · Vals AIAgentic44.9% ±4.524 / 66
Artificial Analysis SciCodeCoding52.1%42 / 114
Artificial Analysis Long Context ReasoningReasoning77.0%82 / 217
Artificial Analysis Omniscience AccuracyKnowledge37.5%63 / 164
Artificial Analysis Humanity's Last ExamKnowledge28.1%91 / 222
Terminal-Bench 2.0 · BenchLMAgentic60.0%17 / 41
Humanity's Last ExamKnowledge41.5%32 / 77
GPQA Diamond · Vals AIKnowledge83.1% ±2.557 / 135
LiveCodeBenchCoding81.5% ±1.160 / 142
MMLU ProKnowledge84.6% ±0.460 / 135
ARC-AGI-2 (semi-private)Reasoning18.9%61 / 131
APEX-Agents-AAAgentic28.2%15 / 30
ARC-AGI-1 (semi-private)Reasoning63.7%64 / 126
MMMU ProMultimodal79.2% ±1.049 / 92
GDPval-AA normalizedAgentic25.6%71 / 133
Vibe Code Bench v1.1Coding48.0% ±5.670 / 123
Artificial Analysis MMMU-ProMultimodal73.3%65 / 113
OSWorld-VerifiedAgentic72.1%22 / 37
FrontierMath-Tiers-1-3-v2-PrivateMath51.2% ±3.054 / 90
Massive Multi-discipline Multimodal Understanding ProMultimodal76.6%32 / 53
SWE-benchCoding73.0% ±2.053 / 87
FrontierMath-Tier-4-2025-07-01-PrivateMath2.1% ±2.135 / 55
LiveBench Data AnalysisReasoning70.8%53 / 80
Terminal-Bench 2.1 · Vals AIAgentic54.7% ±4.651 / 75
IOI v1Coding6.4% ±3.642 / 61
ToolathlonAgentic42.9%16 / 22
MCP AtlasAgentic57.7%39 / 52
Chess PuzzlesReasoning17.0% ±3.870 / 90
SimpleQA VerifiedKnowledge29.4% ±1.471 / 91
Code MigrationCoding12.9% ±3.661 / 77
Humanity's Last Exam without toolsKnowledge28.2%43 / 53
FrontierMath-Tier-4-v2-PrivateMath9.8% ±4.761 / 72
LiveBench CodingCoding71.6%56 / 66
Mystery Game PuzzlesReasoning7.0% ±2.669 / 80
LiveBench Instruction FollowingInstruction59.8%57 / 66
FrontierCode 1.1 MainCoding27.0%15 / 17
Terminal-Bench 4.0Agentic2.5% ±1.054 / 61
LiveBench LanguageKnowledge71.0%59 / 66
LiveBench Agentic CodingAgentic41.7%59 / 66
LiveBench ReasoningReasoning71.3%73 / 80
LiveBench MathematicsMath78.5%61 / 66
Artificial Analysis ITBench-AAAgentic35.2%24 / 24
FrontierMath-2025-02-28-PublicMath50.0% ±16.716 / 64
Finance Agent v1.1Finance53.4% ±2.820 / 51
ARC-AGI-2 (public eval)Reasoning17.8%55 / 122
CorpFin v2Finance60.9% ±1.062 / 131
ARC-AGI-1 (public eval)Reasoning75.1%55 / 116
ProgramBenchCoding0.0%34 / 65
MortgageTaxFinance63.5% ±0.952 / 97
TaxEval v2Finance71.2% ±0.980 / 143
Finance Agent (v2)Finance45.4% ±0.455 / 79
EMBFinance45.4% ±3.658 / 75
Tax Agent BenchFinance8.8% ±2.069 / 82
Legal Research BenchLegal12.5% ±2.378 / 90
CaseLaw v2Legal51.7% ±0.946 / 53
Artificial Analysis Global-MMLU-LiteMultilingual89.4%—
MMMU-Pro with PythonMultimodal78.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYLiveBenchNo licence statedEuroEvalMITARC Prize FoundationTerms restrict reusemodels.devMIT

History

5 releases
Index55.4 → 54.5
23 Oct 202612 Nov 2026
Output / 1M$4.50 → $4.50
23 Oct 202612 Nov 2026

Questions

How much does GPT-5.4 mini cost?
$0.75 per 1M input tokens, $4.50 per 1M output tokens.
How good is GPT-5.4 mini?
Its index is 54.5 out of 100. That is place 108 of 283 ranked models.
What is GPT-5.4 mini best at?
Knowledge: 53.7 out of 100, place 106 of 268.
How fast is GPT-5.4 mini?
95 output tokens per second, 1.3 s to the first token.
What is the context window of GPT-5.4 mini?
400K tokens.
Is GPT-5.4 mini open source?
No. Its weights are not published.
When was GPT-5.4 mini released?
March 2026.

Badge

GPT-5.4 mini on Inference360: place 108 of 283, index 54.5
HTML
<a href="https://inference360.ai/models/gpt-5-4-mini"><img src="https://inference360.ai/badges/gpt-5-4-mini.svg" alt="GPT-5.4 mini on Inference360: place 108 of 283, index 54.5" height="20"></a>
Markdown
[![GPT-5.4 mini on Inference360: place 108 of 283, index 54.5](https://inference360.ai/badges/gpt-5-4-mini.svg)](https://inference360.ai/models/gpt-5-4-mini)

Compare with

Qwen3.6 Plus54.6Inkling54.8Qwen3.5 397B53.8GPT-5.1-Codex-Max53.8GPT-5.4 nano50.9

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeGLM-5.3-Flash63.1 · $0.078/1M

More from OpenAI

All models
GPT-5.4 nano50.9GPT-5.4 Pro73.3GPT-5.464.8GPT-5.3 InstantUnrankedGPT Realtime 1.5Unranked