Inference
EN
OpenAI

GPT-5.2

GPT-5.2 is a reasoning model from OpenAI. 51 benchmarks count toward its index, in 8 categories.

  • Closed weights
  • Reasoning
  • Input: file · image · text
  • Released 2025-12
RankedEnough results for an index.
IndexAll categories, out of 100.59.0±2.3Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.45/sMedian 67/s
First tokenSeconds to the first output token.2.5 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$1.75batch $0.875Median $0.60
Output / 1MUS dollars per 1M output tokens.$14.00batch $7.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.400KMedian 262K
EloLMArena rating and rank.1412±3 #139Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-5.2
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.
$0.175
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
5 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
DatabricksCloud$1.75$14.00400K
DigitalOceanCloud$1.75$14.00400K
OpenRouterRouter$1.75$14.00400K
SAP AI CoreCloud$1.75$14.00400K
Snowflake CortexCloud$1.75$14.00400K

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.9 counted55.3102 / 218
CodingCode writing and repair.7 counted61.065 / 209
ReasoningLogic problems and puzzles.9 counted63.851 / 227
MultimodalTasks with images and text.9 counted51.565 / 84
KnowledgeFacts and expert knowledge.6 counted60.662 / 268
MultilingualTasks in many languages.1 counted51.7—
InstructionTasks with strict prompt rules.2 counted55.524 / 63
MathMath problems.8 counted62.639 / 168

Results

51 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
American Invitational Mathematics Examination 2025Math100.0%1 / 53
AIMEMath96.9% ±0.22 / 95
Terminal-Bench 1.0Agentic63.8% ±5.41 / 47
Massive Multi-discipline Multimodal UnderstandingMultimodal86.7%2 / 28
τ²-bench AirlineAgentic83.0%2 / 22
FrontierMath-2025-02-28-PrivateMath40.7% ±2.97 / 69
Artificial Analysis Long Context ReasoningReasoning82.7%23 / 217
SWE-bench Verified · SWE-bench/experimentsCoding72.8%7 / 65
Graduate-Level Google-Proof Q&AKnowledge92.4%16 / 141
EuroEval SwedishMultilingual63.5% ±0.624 / 195
EuroEval PortugueseMultilingual62.5% ±0.423 / 182
IOI v1Coding54.8% ±8.38 / 61
GPQA Diamond · Vals AIKnowledge91.7% ±1.818 / 135
EuroEval PolishMultilingual57.1% ±0.430 / 209
GPQA diamond · Epoch AIKnowledge91.4% ±1.832 / 217
Artificial Analysis IFBenchInstruction75.4%23 / 146
EuroEval ItalianMultilingual60.1% ±0.428 / 174
Artificial Analysis GPQA DiamondKnowledge90.3%34 / 199
EuroEval SpanishMultilingual53.5% ±0.431 / 176
EuroEval FrenchMultilingual62.0% ±0.432 / 178
FrontierMath-Tier-4-2025-07-01-PrivateMath18.8% ±5.610 / 55
MMMU ProMultimodal86.7% ±0.817 / 92
OTIS Mock AIME 2024-2025Math96.1% ±2.737 / 200
Software Engineering Benchmark VerifiedCoding80.0%22 / 114
EuroEval GermanMultilingual49.6% ±0.540 / 191
Chess PuzzlesReasoning48.0% ±5.019 / 90
LiveCodeBenchCoding85.4% ±1.032 / 142
EuroEval DutchMultilingual56.2% ±0.540 / 175
Artificial Analysis Omniscience AccuracyKnowledge44.3%40 / 164
Artificial Analysis Humanity's Last ExamKnowledge37.7%56 / 222
τ²-bench RetailAgentic81.6%6 / 23
Artificial Analysis MMLU-ProKnowledge87.4%4 / 15
Terminal-Bench 2.0 · Vals AIAgentic51.7% ±5.318 / 66
Critical Physics TasksReasoning11.6%59 / 214
Massive Multi-discipline Multimodal Understanding ProMultimodal79.5%16 / 53
LiveBench MathematicsMath93.2%21 / 66
LiveBench Data AnalysisReasoning78.2%26 / 80
MMLU ProKnowledge86.2% ±0.345 / 135
FrontierMath-Tiers-1-3-v2-PrivateMath67.4% ±2.833 / 90
ARC-AGI-1 (semi-private)Reasoning86.2%47 / 126
ARC-AGI-2 (semi-private)Reasoning52.9%49 / 131
τ²-Bench Tool-Agent-User EvaluationAgentic84.8%58 / 149
FrontierMath-Tier-4-v2-PrivateMath31.7% ±7.332 / 72
LiveBench LanguageKnowledge79.8%30 / 66
Mystery Game PuzzlesReasoning23.0% ±4.239 / 80
SWE-benchCoding75.8% ±1.943 / 87
RealWorldQAMultimodal83.3%9 / 18
τ²-bench BankingAgentic32.2%14 / 28
CharXiv ReasoningMultimodal82.1%25 / 49
Vibe Code Bench v1.1Coding53.5% ±5.163 / 123
Terminal-Bench HardAgentic47.0%12 / 23
SWE-Bench verified · Epoch AICoding73.8% ±2.017 / 32
Gert Labs Composite Game BenchmarkAgentic46.5%28 / 52
τ²-bench TelecomAgentic89.7%10 / 18
Furniture AssemblyReasoning38.3% ±6.321 / 37
LiveBench ReasoningReasoning83.2%47 / 80
CountBenchMultimodal91.9%6 / 10
JobBenchAgentic34.3%22 / 36
EBR-benchReasoning23.0%19 / 31
SimpleQA VerifiedKnowledge37.1% ±1.556 / 91
LiveBench Agentic CodingAgentic50.3%42 / 66
SWE-bench ProCoding55.6%53 / 83
LiveBench CodingCoding76.1%44 / 66
MirrorCodeCoding32.2% ±10.028 / 39
BrowseCompAgentic65.8%48 / 63
ERQAMultimodal59.8%14 / 18
MathVisionMultimodal83.0%19 / 24
LiveBench Instruction FollowingInstruction61.8%54 / 66
SimpleVQAMultimodal55.8%16 / 18
V*Multimodal75.9%13 / 14
OSWorld-VerifiedAgentic47.3%36 / 37
TaxEval v2Finance75.8% ±0.811 / 143
MGSMMultilingual94.0% ±1.46 / 74
FrontierMath-Tier-4-2025-07-01-PublicMath0.0%3 / 36
CaseLaw v2Legal66.0% ±0.68 / 53
FrontierMath-2025-02-28-PublicMath60.0% ±16.311 / 64
Finance Agent v1.1Finance58.5% ±2.99 / 51
MedCodeHealthcare49.7% ±2.320 / 106
CorpFin v2Finance65.9% ±0.925 / 131
MedQAHealthcare94.1% ±0.219 / 95
ARC-AGI-1 (public eval)Reasoning95.0%33 / 116
MortgageTaxFinance67.1% ±0.930 / 97
MedScribeHealthcare84.4% ±1.936 / 109
ARC-AGI-2 (public eval)Reasoning59.8%41 / 122
LegalBenchLegal82.8% ±0.465 / 149
SWE-bench Multilingual · SWE-bench/experimentsCoding66.7%10 / 14
SWE-bench Multilingual · swebench.comMultilingual66.7%9 / 12
Artificial Analysis AIME 2025Math99.0%—
AI2D test splitMultimodal92.2%—
DynaMathMultimodal86.8%—
Video-MME with subtitleMultimodal86.0%—
VideoMMMUMultimodal85.9%—
Video-MME without subtitleMultimodal85.8%—
OmniDocBench 1.5Multimodal85.7%—
MLVU mean averageMultimodal85.6%—
We-MathMultimodal79.0%—
CC-OCRMultimodal70.3%—
LifeVal: everyday life tasksAgentic49.3%—
BioBasicsBench: everyday wet-lab scienceReasoning31.2%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYLiveBenchNo licence statedSierra ResearchMITARC Prize FoundationTerms restrict reuseSWE-bench teamNo licence statedEuroEvalMITmodels.devMIT

History

5 releases
Index60.2 → 59.0
23 Oct 202612 Nov 2026
Output / 1M$14.00 → $14.00
23 Oct 202612 Nov 2026

Questions

How much does GPT-5.2 cost?
$1.75 per 1M input tokens, $14.00 per 1M output tokens.
How good is GPT-5.2?
Its index is 59.0 out of 100. That is place 79 of 283 ranked models.
What is GPT-5.2 best at?
Reasoning: 63.8 out of 100, place 51 of 227.
How fast is GPT-5.2?
45 output tokens per second, 2.5 s to the first token.
What is the context window of GPT-5.2?
400K tokens.
Is GPT-5.2 open source?
No. Its weights are not published.
When was GPT-5.2 released?
December 2025.

Badge

GPT-5.2 on Inference360: place 79 of 283, index 59.0
HTML
<a href="https://inference360.ai/models/gpt-5-2"><img src="https://inference360.ai/badges/gpt-5-2.svg" alt="GPT-5.2 on Inference360: place 79 of 283, index 59.0" height="20"></a>
Markdown
[![GPT-5.2 on Inference360: place 79 of 283, index 59.0](https://inference360.ai/badges/gpt-5-2.svg)](https://inference360.ai/models/gpt-5-2)

Compare with

Qwen3.8-27B59.2Apodex 1.158.9Beam 1.0 501B-A23B FP859.2Beam 1.058.8GPT-5.2-Codex59.9

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-5.2 InstantUnrankedGPT-5.2 ProUnrankedGPT-5.2 ChatUnrankedGPT-5.2-Codex59.9GPT-5.1-Codex-Max53.8