0

GPT-5 Mini

Closed

GPT-5 mini is an AI model from OpenAI.

Publisher
OpenAI
Family
GPT
Context
400K
Input
Output
License
proprietary
Released
7 Aug 2025

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores
prime-hubfrontiermathAALiveBenchSWE-benchVals AI
Attribution policy →
Context
400K
Input
$0.25
/1M tok
Output
$2
/1M tok
Blended
$0.69
#275/ 414/1M · 3:1
Speed
86
#128/ 414tok/s
Cache read
$0.02
0.10x input

Price & speed via Artificial Analysis · ranked across tracked models · caching via OpenRouter

Reported on 63 evals across 12 domains - best 100.0% on Gpu Puzzles Modal · #1 of 3

Reported eval scores

63
Vf Thick Fog272.1
Vf Binary Liar39.2
Gpu Puzzles Modal100
Csv Qa100
Deepdive66.7
Clockbench40
Mastermind1.5
Mastermind1.5
Fingpt Sentiment88.3
Wiki Search86
Wiki Search86
Codenames1.7
Twenty Questions0.5
Hypothesis Forge47.2
Transformerpuzzles0
Backend Bench0
Autonomous Skill Evolution1.1
Curvebench Env18.1
Gutenberg Env90
Capa25
Blackjack Env-0.2
Curvebench Hard Env2.8
Skill Reward Hacking5
Mini Swe Agent Bench80
Trading Bot Bench9.5
Arc Agi Tool0
Text Quests0
Fh Aviary0
LLM Training Puzzles0
DABstep100
BrowseComp20
τ²-bench (Tau²-bench)31.9
TaxEval v275.2
MortgageTax66.9
CorpFin v260.2
IFEval80
LiveBench69.1
MMLU-Pro77.5
Humanity's Last Exam (HLE)5
LiveBench - Language75.5
IFBench45.6
LiveBench - Coding68.2
Terminal-Bench (Hard)14.4
LiveBench - Instruction Following65.3
LiveBench - Math82.2
LiveBench - Data Analysis55.2
LiveBench - Reasoning68.3
SWE-bench Multilingual39.7
SciCode36.9
GPQA Diamond68.7
MedScribe80.6
MedCode43
FrontierMath27.2
LiveCodeBench54.5
SWE-bench Verified59.8
Aider Polyglot Benchmark0
AIME 2025: Problems from the American Invitational Mathematics Examination46.7
ProofBench9
SAGE43
Vibe Code Bench v1.114.2