modelbenchmark.io

GPT-5.4 mini

OpenAI · gpt-5-4-mini

Compare

Strong small GPT for coding subagents, quick tool use, and high-volume work

Specification

openai listing

Context
400K
Max output
128K
Released
2026-03-17
Knowledge cutoff
2025-08-31
Retires
2027-09-21
Open weights
no
Input
text, image
Output
text

Price

US dollars per million tokens · openai list price

Input
$0.75
Output
$4.50
Cache read
$0.075
Cache write
$0.75

Batch and priority prices

US dollars per million tokens · per listing

PriceListing$/M
Batch inputopenai$0.375
Batch outputopenai$2.25
Priority inputazure_ai$1.50
Priority outputazure_ai$9.00

Batch input is 50% below the interactive input price at the same listing.

Quality

10 benchmarks · 2 sources

Composite
50th
percentile of 334 scored models
Rank
169 / 334
±0.07 sd
Evidence
10 × 2
3 or more benchmarks, from 2 or more sources
Effort range
3.3
points between effort settings
reasoning67th3/4 bench
math58th5/7 bench
knowledge24th1/1 bench
general7th1/1 bench

By reasoning effort

every setting placed on the same scale as the leaderboard

SettingCompositeBenchmarks behind it
high69th6
xhigh63rd6
low25th2
medium18th1

Every score

one row per source and configuration — nothing averaged away

BenchmarkScoreConfigurationSourceRun
OTIS Mock AIME 2024-202588.9 ±4.7xhighEpoch AI2026-08-07
87.2 ±4.3highEpoch AI2026-04-15
Settings differ by 1.7 points on this benchmark.
GPQA diamond86.9 ±2.4xhighEpoch AI2026-08-07
83.6 ±2.2highEpoch AI2026-04-15
Settings differ by 3.3 points on this benchmark.
FrontierMath-v139.1 ±16.7high · 2 runsEpoch AI2026-04-15
Chess Puzzles24.0 ±4.3xhighEpoch AI2026-08-07
18.0 ±3.9highEpoch AI2026-04-15
Settings differ by 6.0 points on this benchmark.
FrontierMath-Tiers-1-3-v251.2 ±3.0xhighEpoch AI2026-06-12
24.6 ±2.6lowEpoch AI2026-08-28
Settings differ by 26.6 points on this benchmark.
FrontierMath-Tier-4-2025-07-012.1 ±2.1highEpoch AI2026-04-15
SimpleQA Verified29.4 ±1.4highEpoch AI2026-08-27
FrontierMath-Tier-4-v29.8 ±4.7xhighEpoch AI2026-06-12
Mystery Game Puzzles8.0 ±2.7lowEpoch AI2026-08-27
7.0 ±2.6mediumEpoch AI2026-08-27
Settings differ by 1.0 points on this benchmark.
LiveBench66.6xhigh · 23 runsLiveBench2026-06-25

The leading figure for a benchmark is the median across its configurations, so one heroic high-effort run cannot set the number.

Available from 38 hosts

HostIn $/MOut $/MCache rdCache wrContextOutputRetires
openaivendor$0.75$4.50$0.075
400K
272K
128K
kenarifreefree400K128K
xpersona$0.375$4.00$0.0375272K128K
poe$0.68$4.00$0.068400K128K
302ai$0.75$4.50400K128K
abacus$0.75$4.50$0.075400K128K
aihubmix$0.75$4.50$0.075400K128K
azure$0.75$4.50$0.075
400K
272K
128K2027-09-21
azure_ai$0.75$4.50$0.075272K128K2027-09-21
azure-cognitive-services$0.75$4.50$0.075400K128K
cloudflare-ai-gateway$0.75$4.50$0.075128K128K
crossmodel$0.75$4.50$0.075$0.75400K128K
databricks$0.75$4.50$0.075$0.75
400K
272K
128K
edenai$0.75$4.50$0.075400K128K
fastrouter$0.75$4.50400K128K
freemodel$0.75$4.50$0.075$0.75400K128K
frogbot$0.75$4.50$0.075400K128K
github-copilot$0.75$4.50$0.075400K128K
impossibl$0.75$4.50$0.075400K128K
kilo$0.75$4.50$0.075400K128K
llmgateway$0.75$4.50$0.075400K128K
llmgateway-providers$0.75$4.50$0.075400K128K
merge-gateway$0.75$4.50$0.075400K128K
nano-gpt$0.75$4.50$0.075400K128K
nearai$0.75$4.50$0.075400K128K
neon$0.75$4.50$0.075400K128K
ofox$0.75$4.50$0.075400K128K
opencode$0.75$4.50$0.075400K128K
openrouter$0.75$4.50$0.075
400K
272K
128K
opper$0.75$4.50$0.075400K128K
orcarouter$0.75$4.50$0.075400K128K
perplexity$0.75$4.50$0.075
pioneer$0.75$4.50$0.075$0.75400K128K
requesty$0.75$4.50$0.075400K128K
vercel$0.75$4.50$0.075400K128K
vivgrid$0.75$4.50$0.075400K128K
zenmux$0.75$4.50400K128K
model-oracle-ai400K128K

Price history

append-only observations · a listing writes a row only when its price moves

DateHostField$/M
2026-09-02openrouterCache read$0.075
2026-09-06openrouterCache read$0.075
2026-09-02openrouterInput$0.75
2026-09-06openrouterInput$0.75
2026-09-02openrouterOutput$4.50
2026-09-06openrouterOutput$4.50

Other listings of this model

same model, different key

More from OpenAI

most-hosted first