Which is better, Claude 4 Opus or DeepSeek-R1?

Across 50 shared benchmarks tracked on Vector Wire, Claude 4 Opus leads on 25 and DeepSeek-R1 leads on 23 — Claude 4 Opus leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude 4 Opus vs DeepSeek-R1

AnthropicvsDeepSeek50 shared benchmarks25–23 head-to-head

Benchmark	Claude 4 Opus	DeepSeek-R1
AA Intelligence	31.7	18.6
AA-LCR	40	56
aider_polyglot	70.7	71.6
AIME 2024	76	91.4
AIME 2025	75.5	87.5
AIR-Bench 2024	85.7	52.9
anthropic_red_team	98.9	99.1
arc_agi_1	27	21.2
ARC-AGI-2	8.6	1.3
Artificial Analysis Coding Index	34	24.6
bbq	99.3	96.6
CNMO 2024	57.6	78.8
Fortress	27.6	74.4
FullStackBench	70.3	70.1
GPQA Diamond	79.6	81
harmbench	91.7	54.6
HLE	12.3	17.7
HMMT 2025	15.9	79.4
hmmt_feb_2025	60	76.7
IFBench	53.7	39
ifeval	87.4	83.3
LiveCodeBench	70.4	84.4
LiveCodeBench (24/8~25/5)	56.6	73.1
livecodebench_easy	99.1	99.2
livecodebench_hard	33.1	63.6
livecodebench_medium	80.4	90.9
longbench_v2	55.6	58.3
math_500_em	98.2	98
mmlu	92.9	92.9
mmlu_redux	94.2	93.4
MMLU-Pro	86.6	85
multichallenge	58.6	45
MultiNRC	33.9	27.6
OpenAI-MRCR (128k)	48.9	51.5
scicode	40.9	35.7
simple_safety_tests	100	98.3
simplebench	58.8	40.8
simpleqa	22.8	92.3
SWE-bench Verified	79.4	57.6
TAU-bench (airline)	59.6	53.5
TAU-bench (retail)	81.4	63.9
Terminal-Bench	43.2	5.7
Terminal-Bench Hard	31.1	6.1
vectara_answer_rate	91	97
vectara_avg_summary_length	123.2	93.5
vectara_factual_consistency	88	88.7
vectara_hallucination_rate ↓	12	11.3
xstest	97	98.8
ZebraLogic	95.1	95.1
τ²-Bench Telecom (AA run)	73.4	11.4

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.