Which is better, Claude Opus 4.5 or Claude Sonnet 4.5?

Across 81 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 72 and Claude Sonnet 4.5 leads on 7 — Claude Opus 4.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.5 vs Claude Sonnet 4.5

AnthropicvsAnthropic81 shared benchmarks72–7 head-to-head

Benchmark	Claude Opus 4.5	Claude Sonnet 4.5
AA Agentic Index	59.6	50.6
AA Intelligence	41.9	37.4
AA-LCR	76	68.3
AA-Omniscience	14	-0.1
AIME 2025	92.8	88
AIME25 no tools	91	88
arc_agi_1	82	63.7
ARC-AGI-2	37.6	13.6
ARC-AGI-2 (Verified)	37.6	13.6
arena_elo	1473	1456
arena_text_factuality	1458	1448
Artificial Analysis Coding Index	47.8	52.1
browsecomp	37	24.1
BrowseComp (context management)	59.2	26.1
browsecomp_zh	62.4	42.4
BrowseComp-Plus (with tool result clearing + memory)	72.9	67.2
BrowseComp-Plus (with tool result clearing)	67.6	60.4
CC-OCR	76.9	68.1
charxiv_rq	68.5	67.2
coding_arena_elo	1495	1392
critpt	4.6	1.1
Fortress	13.6	17
frontiermath_tier_4	4.2	4.2
gdpval	47.3	40.4
GPQA Diamond	87	83.4
HLE	30.8	19.8
HLE (with tools)	43.4	32
hmmt_feb_2025	92.9	79.2
hmmt_nov_2025	93.3	81.7
IFBench	58	57.3
imo_answer_bench	84	65.9
LAB-Bench FigQA (no tools)	54.9	52.3
LAB-Bench FigQA (with tools)	69.2	63.7
LiveCodeBench	87	71
LiveCodeBench v6	84.8	64
longbench_v2	64.4	61.8
MathVista	80.2	79.8
MCP Atlas	69.8	59.5
mmlu_redux	95.6	95.6
MMLU-Pro	90	88.2
mmmlu	90.8	89.1
MMMU	80.7	79.6
MMMU-Pro	74	68.7
Multi-SWE-Bench	50	44.3
multichallenge	59	55.3
MultiNRC	48.6	35.8
OctoCodingbench	36.2	22.8
OJ-Bench (cpp)	54.6	30.4
OmniDocBench 1.5	87.7	85.8
OmniScience Accuracy	46.6	32.9
OmniScience Non-Hallucination	39	50.9
OSWorld-Verified	66.3	61.4
RealWorldQA	77	70.3
scicode	50	45
SEAL VISTA	46.4	48.8
Seal-0	47.7	53.4
simplebench	62	54.3
simpleqa_verified	41.8	23.6
SimpleVQA	69.7	57.6
swe_bench_bash	76.8	71.4
SWE-bench Multilingual	77.5	68
SWE-bench Pro	57.1	43.6
SWE-bench Verified	81.5	82
SWE-Perf	4.7	3
SWT-bench	80.2	69.5
Terminal-Bench 2.0	59.3	50
Terminal-Bench Hard	47	35.6
toolathlon	43.5	41
vectara_answer_rate	98.7	95.6
vectara_avg_summary_length	114.5	127.8
vectara_factual_consistency	89.1	88
vectara_hallucination_rate ↓	10.9	12
VIBE (Average)	90.7	85.2
VIBE-Android	92.2	87.5
VIBE-Backend	98	90.8
VIBE-iOS	90	81.2
VIBE-Simulation	84	79.1
VIBE-Web	89.1	87.3
τ²-Bench	98.2	98
τ²-Bench (Retail)	88.9	86.2
τ²-Bench Telecom (AA run)	89.5	78.1

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.