OpenAI deep research

OpenAI deep research ranks first on 3 of the 26 benchmarks it has scores on.

OpenAI Released 2 Feb 2025 Family o3 The lab's model card

Benchmarks with a score

Ranked first, out of five or more models

In the top three

OpenAI Feb 2025

Closed weights Flagship

No Epoch Capabilities Index score yet

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Benchmark	Score		Rank	Source
PersonQA	86%		#2 of 9	Model card
Humanity's Last Exam	26.6%		#81 of 147	Model card

Benchmark	Score	Rank	Source
Agentic tasks (autonomous replication and adaptation)	78%	#1 of 4	Model card
SWE-Lancer Diamond (SWE Manager)	51%	#1 of 4	Model card
OpenAI PRs	42%	#6 of 12	Model card
SWE-Lancer	53%	#5 of 10	Model card
SWE-bench Verified	68%	#88 of 150	Model card
OpenAI Research Engineer Interview: Multiple Choice	78%	#6 of 9	Model card
MLE-bench	11%	#14 of 18	Model card
OpenAI Research Engineer Interview: Coding	79%	#6 of 7	Model card

Benchmark	Score	Rank	Source
SnakeBench	30	#39 of 183	Official leaderboard
DeepSearchQA	44.2%	#5 of 10	Official leaderboard
DeepResearch Bench	46.5%	#2 of 4	Official leaderboard
LiveResearchBench	62.9	#4 of 7	Official leaderboard
GAIA	67.4%	#5 of 5	Model card
DeepResearch Bench II	43.1%	#1 of 1	Official leaderboard

Benchmark	Score	Rank	Source
BioLP-bench	49%	#1 of 9	Model card
Capture the Flag (High School)	82%	#1 of 9	Model card
WMDP Biology	90%	#1 of 7	Model card
Capture the Flag (Collegiate)	55%	#2 of 13	Model card
Contextual Nuclear Knowledge	77%	#1 of 4	Model card
Radiological and Nuclear Expert Knowledge	74%	#1 of 4	Model card
MakeMePay	36%	#2 of 6	Model card
Long-form Biological Risk Questions	0%	#1 of 3	Model card
Multimodal Troubleshooting Virology	55%	#9 of 26	Model card
ChangeMyView	84.1%	#2 of 5	Model card
Tacit Knowledge and Troubleshooting	72%	#13 of 29	Model card
StrongREJECT	93%	#10 of 22	Model card
Capture the Flag (Professional)	47%	#10 of 21	Model card
Challenging Red Teaming Evaluation 2 (created for deep research)	67%	#1 of 2	Model card
MakeMeSay	24%	#6 of 9	Model card
ProtocolQA Open-Ended	28%	#19 of 25	Model card
Standard Refusal Evaluation	97%	#7 of 8	Model card
Challenging Refusal Evaluation	89%	#8 of 9	Model card
Challenging Red Teaming Evaluation 1 (created for o3-mini)	50%	#2 of 2	Model card
Red Teaming Network pairwise safety comparison	59.2%	#2 of 2	Model card

The benchmarks the lab chose to headline, with the scores it gave.

Every result in the longer documents the lab published with it. 32 of them aren't in the launch post.