GPT-5
Aug 2025
GPT-5Compare inon
Benchmarks of GPT-569
BenchmarkByScoreRank
Overall1
Knowledge4
Reasoning4
Math15
Science2
Coding5
Agents2
Multimodal2
Writing17
Health & law3
Documents1
Legacy · saturated or no longer reported13
No benchmark matches
Rankingat high effort?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperGemini 3.8 Flashhighwins 65%-56%*$3.49Sampling parameters are ignored+1BetterGPT-5.6 Terraxhighwins 57%+31%$0.63Prompts with >272K input tokens
OpenKimi K3maxwins 63%+75%$2.00Sampling parameters are fixed+3CheaperGemini 3 Flash Previewunstatedwins 62%-67%$0.50 / $3CheaperGemini 3.6 Flashhighwins 60%-56%*$1.60ModelScoreCostPer taskWhat changes
Gemini 3.8 Flashhighwins 65%-56%*$3.49Sampling parameters are ignored+1Gemini 3 Flash Previewunstatedwins 62%-67%$0.50 / $3Gemini 3.6 Flashhighwins 60%-56%*$1.60
Muse Spark 1.3maxwins 59%-42%$3.69Only auto tool_choice is supported+5
GLM 5.3maxOpenwins 58%-37%$2.01No image input+4ModelScoreCostPer taskWhat changes
GPT-5.6 Terraxhighwins 57%+31%$0.63Prompts with >272K input tokensGPT-5.6 Solxhighwins 60%+133%$1.18GLM 5.2maxOpenwins 57%-37%$1.28No image inputGPT-5.5highwins 64%+227%$1.40Prompts with >272K input tokensGemini 3.6 Flashhighwins 60%-56%*$1.60ModelScoreCostPer taskWhat changes
Kimi K3maxOpenwins 63%+75%$2.00Sampling parameters are fixed+3GLM 5.3maxOpenwins 58%-37%$2.01No image input+4GLM 5.2maxOpenwins 57%-37%$1.28No image input
DeepSeek V4 Pro 0813unstatedOpenwins 56%-42%$1.32 / $3.96No image input+5Kimi K2.6unstatedOpenwins 56%-50%$0.95 / $4Context: 400K → 256KModelScoreCostPer taskWhat changes
GPT-5.5highwins 64%—$1.40Prompts with >272K input tokensGPT-5.4highwins 62%—$2.50 / $15GPT-5.6 Solxhighwins 60%—$1.18GPT-5.1highwins 57%—$1.25 / $10GPT-5.6 Terraxhighwins 57%—$0.63Prompts with >272K input tokensGPT-5.2highwins 53%—$1.75 / $14GPT-5.6 Lunaxhighsame—$0.085GPT-6 Astramaxsame—$3.26Monitoring can return HTTP 403+2All OpenAI models →History
17 versionsBenchmarks
all categoriesChat & Knowledge
Humanity's Last Exam28.5#90
Arena Text · overall1,435#70
Arena Text · Chinese1,421.8#104
Arena Text · English1,409.1#99
Arena Text · French1,407.9#96
Arena Text · German1,415.9#69
Arena Text · instruction following1,388.3#94
Arena Text · Japanese1,402.4#49
Arena Text · Korean1,358.6#80
Arena Text · non-English1,397#89
Arena Text · Polish1,411.6#68
Arena Text · Russian1,401.9#84
Arena Text · Spanish1,388#101
AA Omniscience Index-8.7#91
CritPt (physics)5.7#77
AA Omniscience · Accuracy40.3#47
AA-LCR (long context) 1.178.2#69
MathArena · BRUMO 202591.7#12
MathArena · CMIMC 202588.8#8
MathArena · IMO 202538.1#1
MMLU-Pro · official boardlab87.1
HealthBench Professional length-adjustedlab46.2
AIME 2025lab94.6
HMMT February 2025lab88.3
HMMT November 2025lab89.2
IMOAnswerBenchlab76
HMMT 2025 chart edition unstatedlab88.3
Arena Text · overall · October 22, 20261,406.2#89
Arena Text · overall · October 15, 20261,406.2#87
GPQA Diamond85.4#79
AIME 2025 · AA archived94.3#7
Coding
Terminal Bench 2.0lab35.2
LiveCodeBenchlab84.5
Codeforces Ratinglab2,537
SWE Verifiedlab74.9
Terminal-Bench 2.135.2#92
Aider Polyglot · two attempts88#1
Agents & Tool Use
GDPval-AA (work tasks) 2.120.9#136
τ³-Banking · AA22.1#48
Tau2-Benchlab80.2
MCP-Universelab47.9
MCP-Marklab50.9
Tool-Decathlonlab29
HLE without toolslab26.3
τ²-bench · AA84.8#64
Writing & Language
Arena Text · creative writing1,372#87
IFBench (instructions)73.1#33
Vision & Documents
MMMU-Pro (vision)74.2#57
Arena Vision · overall1,208.2#62
FAQ
How much does GPT-5 cost?›
$1.25 per 1M input tokens and $10 per 1M output tokens; cached input $0.13 per 1M.
Is GPT-5 free to use?›
There is no free API; the cheapest paid price is $1.25 per 1M input tokens.
How good is GPT-5, and what is it best at?›
#90 of 336 in Chat & Knowledge, #8 of 19 in Coding, #136 of 177 in Agents & Tool Use, #88 of 197 in Writing & Language, #57 of 125 in Vision & Documents. Strongest at: Long context, Instructions.
Is GPT-5 better than o3?›
On Arena Text · Japanese, GPT-5 scores 1,402.4 vs 1,403.1 — about the same.
How can I use GPT-5?›
Through the API of Azure, OpenAI.
What is the context window of GPT-5?›
400K tokens, with up to 128K output tokens.
When was GPT-5 released? Is there a newer version?›
Released 2025-08-07. It has been superseded by GPT-5.6 Sol.
Is GPT-5 open source?›
No, GPT-5 is a closed model available through an API. The closest open-weights option here is Kimi K3.
Can GPT-5 read images?›
It accepts text and images, PDF files as input.
What is GPT-5?›
GPT-5 is a closed language model from OpenAI, released Aug 2025.
What are the best alternatives to GPT-5?›
Gemini 3.8 Flash (cheaper), GPT-5.6 Terra (better), Kimi K3 (open weights).