Which is better, DeepSeek-R1 or o4-mini?

Across 45 shared benchmarks tracked on Vector Wire, DeepSeek-R1 leads on 18 and o4-mini leads on 26 — o4-mini 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 o4-mini

DeepSeekvsOpenAI45 shared benchmarks18–26 head-to-head

Benchmark	DeepSeek-R1	o4-mini
AA Agentic Index	3.1	36.1
AA Intelligence	18.6	26.1
AA-LCR	56	60
AA-Omniscience	-31.3	-35.7
aider_polyglot	71.6	72
AIME 2025	87.5	92.7
AIR-Bench 2024	52.9	78.5
anthropic_red_team	99.1	98.2
arc_agi_1	21.2	58.7
ARC-AGI-2	1.3	6.1
Artificial Analysis Coding Index	24.6	25.6
bbq	96.6	94
browsecomp	8.9	51.5
critpt	0.6	0.6
Fortress	74.4	21.5
gdpval	1.5	25.4
GPQA Diamond	81	81.4
harmbench	54.6	97
HLE	17.7	16.5
IFBench	39	68.7
imo_2025	6.8	14.3
LiveCodeBench	84.4	84.5
livecodebench_easy	99.2	98.8
livecodebench_hard	63.6	62.9
livecodebench_medium	90.9	92.2
mmlu_prox	75.5	69.3
multichallenge	45	44.9
MultiNRC	27.6	22.2
OmniScience Accuracy	30.7	24.8
OmniScience Non-Hallucination	10.5	19.5
scicode	35.7	46.5
simple_safety_tests	98.3	100
simplebench	40.8	38.7
simpleqa_verified	27.4	23.9
SWE-bench Verified	57.6	68.1
TAU-bench (airline)	53.5	49.2
TAU-bench (retail)	63.9	71.8
Terminal-Bench Hard	6.1	15.2
usamo_2025	30.1	19.1
vectara_answer_rate	97	99.2
vectara_avg_summary_length	93.5	130.9
vectara_factual_consistency	88.7	81.4
vectara_hallucination_rate ↓	11.3	18.6
xstest	98.8	97.4
τ²-Bench Telecom (AA run)	11.4	55.6

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.