Which is better, Claude 4 Opus or o3?

Across 49 shared benchmarks tracked on Vector Wire, Claude 4 Opus leads on 18 and o3 leads on 29 — o3 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 o3

AnthropicvsOpenAI49 shared benchmarks18–29 head-to-head

Benchmark	Claude 4 Opus	o3
AA Intelligence	31.7	31.1
AA-LCR	40	73.3
aider_polyglot	70.7	81.3
AIME 2024	76	91.6
AIME 2025	75.5	89.2
AIR-Bench 2024	85.7	84.5
anthropic_red_team	98.9	98.3
arc_agi_1	27	60.8
ARC-AGI-2	8.6	6.5
arena_vision	1207	1217
Artificial Analysis Coding Index	34	38.4
bbq	99.3	97.9
Fortress	27.6	16
frontiermath_tier_4	4.2	2.1
FullStackBench	70.3	69.3
GPQA Diamond	79.6	83.3
harmbench	91.7	98.4
HLE	12.3	20.3
hmmt_feb_2025	60	77.5
IFBench	53.7	71.4
LiveCodeBench	70.4	84.7
LiveCodeBench (24/8~25/5)	56.6	75.8
livecodebench_easy	99.1	99.1
livecodebench_hard	33.1	66
livecodebench_medium	80.4	89.8
longbench_v2	55.6	58.8
math_500_em	98.2	98.1
metr_hcast	61.7	64.7
metr_re_bench	20	15
metr_swaa	99.8	99.8
MMLU-Pro	86.6	85
MMMU	73.7	82.9
multichallenge	58.6	60.4
MultiNRC	33.9	45.5
OpenAI-MRCR (128k)	48.9	56.5
scicode	40.9	41
simple_safety_tests	100	99
simplebench	58.8	53.1
simpleqa	22.8	49.4
swe_bench_bash	67.6	58.4
SWE-bench Verified	79.4	69.1
TAU-bench (airline)	59.6	52
TAU-bench (retail)	81.4	73.9
Tau2 airline	60	64.8
Tau2 retail	81.8	80.2
Terminal-Bench Hard	31.1	37.1
xstest	97	97.3
ZebraLogic	95.1	95.8
τ²-Bench Telecom (AA run)	73.4	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.