sugra.blog

BlogLLM Benchmark Scores as Published, Side by Side

GPT-5 mini: Benchmark Scores as Published

Every result the catalog holds for GPT-5 mini: 83 tests in 12 areas, from 3 pages. 16 printed by OpenAI itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.

Developer OpenAIReleased 2025-08-07Context window 400KMax output 128KInput Text, imageOutput TextWeights ClosedData version 2026103001, as of 2026-10-30

Compare in the table

Agentic Coding

TestValuePrinted bySetup as printed
SWE-bench Verified72%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
SWE-Bench Pro45.7%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high
Terminal-Bench 2.038.2%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15; printed on 1 more pageEffort: high

Science and Math

TestValuePrinted bySetup as printed
GPQA Diamond81.6%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15; printed on 1 more pageEffort: high
HMMT February 202589.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
HMMT November 202584.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
PolyMATH67.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Long Context

TestValuePrinted bySetup as printed
AA-LCR68%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
LongBench v256.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
GraphWalks BFS up to 128k (F1)73.4%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high; Context: 0K-128K
MRCR v2 8-needle 128K-256K19.4%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high; Context: 128K-256K
MRCR v2 8-needle 64K-128K35.1%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high; Context: 64K-128K
GraphWalks parents up to 128k (accuracy)64.3%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high; Context: 0-128K

Computer Use

TestValuePrinted bySetup as printed
OSWorld-Verified42%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high

Multimodal Understanding

TestValuePrinted bySetup as printed
CharXiv Reasoning (no tools)68.6%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMMU-Pro67.5%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15; printed on 1 more pageEffort: high
MMMU (validation)79%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MathVision71.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MathVista (testmini)79.1%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
AI2D (test)88.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMMU-Pro (with tools)74.1%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high; Tools: python
OmniDocBench 1.577%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Video-MMMU82.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
OCRBench82.1%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
HallusionBench63.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMStar74.1%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
BabyVision (no tools)20.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
ERQA54%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
SimpleVQA56.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Video-MME (with subtitles)83.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
RealWorldQA79%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMVU69.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
V* Bench (tool setting not stated)71.7%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
CC-OCR70.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMBench-EN dev v1.186.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
CountBench91%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
DynaMath81.4%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
ZeroBench (subquestions)27.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
EmbSpatial-Bench80.7%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MLVU83.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMLongBench-Doc50.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Video-MME (without subtitles)78.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
ZeroBench (metric not stated)3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
RefSpatial-Bench9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
OmniDocBench 1.5 (overall edit distance)lower is better0.1791OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: none; Tools: none
LingoQA62.4%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
VLMs are Blind75.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
TIR-Bench (tool setting not stated)24.6%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

General Knowledge and Reasoning

TestValuePrinted bySetup as printed
Humanity's Last Exam (no tools)18.3%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15; printed on 1 more pageEffort: high; Tools: none
MMLU-Pro83.7%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Humanity's Last Exam (with tools)31.6%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15; printed on 1 more pageEffort: high; Tools: with tools
MMLU-Redux93.7%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
C-Eval82.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
SuperGPQA58.6%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Code Generation

TestValuePrinted bySetup as printed
LiveCodeBench v680.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Codeforces rating2160Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
OJBench (language not stated)40.4%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
FullStackBench en30.6%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
FullStackBench zh35.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Tool Use and Workflows

TestValuePrinted bySetup as printed
Toolathlon26.9%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high
τ²-bench Telecom74.1%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high
MCP Atlas47.6%OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15Effort: high
τ²-bench (domain not stated)69.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
DeepPlanning17.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
VitaBench13.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
BFCL v455.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Web Search and Research

TestValuePrinted bySetup as printed
BrowseComp48.1%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15Tools: enabled; search; Context: 256000
BrowseComp-ZH49.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15Tools: enabled; search; Context: 256000
WideSearch (metric not stated)47.2Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15Tools: enabled; search; Context: 256000
Seal-034.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15Tools: enabled; search; Context: 256000

Instruction Following

TestValuePrinted bySetup as printed
IFBench75.4Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MultiChallenge59%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
IFEval93.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Multilingual

TestValuePrinted bySetup as printed
MMMLU86.2%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MMLU-ProX78.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
WMT24++ (XCOMET-XXL)80.7Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
Global PIQA88.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
INCLUDE81.8%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
MaXIFE85.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
NOVA-6351.9%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Healthcare

TestValuePrinted bySetup as printed
MedXpertQA MM34.4%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
PMC-VQA36.3%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-
SLAKE70.5%Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15-

Price

API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked.

PriceValuePrinted bySetup as printed
Input$0.25OpenAI, OpenAI API pricingchecked 2026-10-15-
Cached input$0.025OpenAI, OpenAI API pricingchecked 2026-10-15-
Output$2OpenAI, OpenAI API pricingchecked 2026-10-15-
Input, batch$0.125OpenAI, OpenAI API pricingchecked 2026-10-15-
Cached input, batch$0.0125OpenAI, OpenAI API pricingchecked 2026-10-15-
Output, batch$1OpenAI, OpenAI API pricingchecked 2026-10-15-
Input, flex$0.125OpenAI, OpenAI API pricingchecked 2026-10-15-
Cached input, flex$0.0125OpenAI, OpenAI API pricingchecked 2026-10-15-
Output, flex$1OpenAI, OpenAI API pricingchecked 2026-10-15-
Input, fast$0.45OpenAI, OpenAI API pricingchecked 2026-10-15-
Cached input, fast$0.045OpenAI, OpenAI API pricingchecked 2026-10-15-
Output, fast$3.6OpenAI, OpenAI API pricingchecked 2026-10-15-

More from OpenAI

Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.