Which is better, DeepSeek-R1 or Gemini 2.5 Pro?

Across 50 shared benchmarks tracked on Vector Wire, DeepSeek-R1 leads on 12 and Gemini 2.5 Pro leads on 38 — Gemini 2.5 Pro 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 Gemini 2.5 Pro

DeepSeekvsGoogle50 shared benchmarks12–38 head-to-head

Benchmark	DeepSeek-R1	Gemini 2.5 Pro
AA Agentic Index	3.1	7.2
AA Intelligence	18.6	27
AA-LCR	56	66
AA-Omniscience	-31.3	-14.3
aider_polyglot	71.6	82.2
AIME 2025	87.5	88.3
AIR-Bench 2024	52.9	73.6
anthropic_red_team	99.1	99.5
arc_agi_1	21.2	41
ARC-AGI-2	1.3	4.9
Artificial Analysis Coding Index	24.6	46.7
bbq	96.6	96.4
browsecomp	8.9	9.9
browsecomp_zh	35.7	32.2
critpt	0.6	2.6
Fortress	74.4	54.9
gdpval	1.5	8.5
GPQA Diamond	81	84.4
harmbench	54.6	65.4
HLE	17.7	22.5
HMMT 2025	79.4	65.7
hmmt_feb_2025	76.7	82.5
IFBench	39	49
imo_2025	6.8	31.6
LiveCodeBench	84.4	82.7
livecodebench_easy	99.2	98.8
livecodebench_hard	63.6	59.4
livecodebench_medium	90.9	90.6
MMLU-Pro	85	86
multichallenge	45	53.6
OmniScience Accuracy	30.7	39
OmniScience Non-Hallucination	10.5	12.6
scicode	35.7	43
simple_safety_tests	98.3	97
simplebench	40.8	62.4
simpleqa	92.3	50.8
simpleqa_verified	27.4	56
SWE-bench Verified	57.6	67.2
TauBench V3 - Banking	6.4	9.7
Terminal-Bench	5.7	25.3
Terminal-Bench 2.1	19.1	28.5
Terminal-Bench Hard	6.1	26.5
usamo_2025	30.1	24.4
vectara_answer_rate	97	99.1
vectara_avg_summary_length	93.5	106.4
vectara_factual_consistency	88.7	93
vectara_hallucination_rate ↓	11.3	7
xstest	98.8	98.7
τ²-Bench Telecom (AA run)	11.4	54.1
τ³-Bench	6.4	9.3

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.