Frontier Log

OpenAIModel

OpenAI o3

Made by
OpenAI
Released
Announced in
Introducing OpenAI o3 and o4-miniopenai.com
Event
o3 and o4-mini
API record
/api/v1/models/openai-o3

Benchmark results

BenchmarkScoreSource check
Humanity's Last ExamNo tools · high reasoning effort20.32%
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
rerun on a later version of HLE; full set; effort not stated · reported later, in Introducing GPT-514.7%
CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
high reasoning effort · reported later, in Introducing GPT-5 for developers20.2%
CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
Humanity's Last ExamWith tools · high reasoning effort; Python and browsing; known cheating domains blocked; attempts a monitor flags count as wrong24.90%
CorroboratedThe same number is printed in 2 other lab documents: 1 2
rerun on a later version of HLE; full set; Python and browser · reported later, in Introducing GPT-524.3%
CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
SWE-bench Verifiedhigh reasoning effort; 477 of the 500 verified tasks; 256K context69.1%
CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 5 other lab documents, the first 3 linked: 1 2 3
GPQA Diamondhigh reasoning effort; no tools83.3%
CorroboratedThe same number is printed in 2 other lab documents: 1 2
BrowseComphigh reasoning effort; Python and browsing49.7%
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
AIME 2024No tools · high reasoning effort91.6%
CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 1 other lab document: 1
AIME 2025No tools · high reasoning effort88.9%
CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 1 other lab document: 1
AIME 2025With tools · Python interpreter; pass@1; OpenAI says not to compare with-tools scores to models without tools98.4%
CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
Python interpreter; majority vote over 8 samples · parallel test-time compute100%
CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
Sources

Reported by OpenAI

Events

Stories

All models