Which is better, DeepSeek-R1 or o3?

Across 49 shared benchmarks tracked on Vector Wire, DeepSeek-R1 leads on 8 and o3 leads on 40 — o3 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-R1 vs o3

DeepSeekvsOpenAI49 shared benchmarks8–40 head-to-head

Benchmark	DeepSeek-R1	o3
AA Agentic Index	3.1	36.1
AA Intelligence	18.6	31.1
AA-LCR	56	73.3
AA-Omniscience	-31.3	-15.3
aider_polyglot	71.6	81.3
AIME 2024	91.4	91.6
AIME 2025	87.5	89.2
AIR-Bench 2024	52.9	84.5
anthropic_red_team	99.1	98.3
arc_agi_1	21.2	60.8
ARC-AGI-2	1.3	6.5
Artificial Analysis Coding Index	24.6	38.4
bbq	96.6	97.9
browsecomp	8.9	49.7
critpt	0.6	1.1
Fortress	74.4	16
FullStackBench	70.1	69.3
gdpval	1.5	12.8
GPQA Diamond	81	83.3
harmbench	54.6	98.4
HLE	17.7	20.3
hmmt_feb_2025	76.7	77.5
IFBench	39	71.4
imo_2025	6.8	16.7
LiveCodeBench	84.4	84.7
LiveCodeBench (24/8~25/5)	73.1	75.8
livecodebench_easy	99.2	99.1
livecodebench_hard	63.6	66
livecodebench_medium	90.9	89.8
longbench_v2	58.3	58.8
math_500_em	98	98.1
MMLU-Pro	85	85
multichallenge	45	60.4
MultiNRC	27.6	45.5
OmniScience Accuracy	30.7	38.6
OmniScience Non-Hallucination	10.5	12.9
OpenAI-MRCR (128k)	51.5	56.5
scicode	35.7	41
simple_safety_tests	98.3	99
simplebench	40.8	53.1
simpleqa	92.3	49.4
simpleqa_verified	27.4	53
SWE-bench Verified	57.6	69.1
TAU-bench (airline)	53.5	52
TAU-bench (retail)	63.9	73.9
Terminal-Bench Hard	6.1	37.1
xstest	98.8	97.3
ZebraLogic	95.1	95.8
τ²-Bench Telecom (AA run)	11.4	80.7

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.