Which is better, Claude Sonnet 5 or Opus 5?

Across 43 shared benchmarks tracked on Vector Wire, Claude Sonnet 5 leads on 3 and Opus 5 leads on 40 — Opus 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 Sonnet 5 vs Opus 5

AnthropicvsAnthropic43 shared benchmarks3–40 head-to-head

Benchmark	Claude Sonnet 5	Opus 5
AA Agentic Index	49.7	59.2
AA Intelligence	55.3	63.1
AA-LCR	77	78.7
AA-Omniscience	16.4	37.1
arena_elo	1463	1487
arena_text_factuality	1453	1481
Artificial Analysis Coding Index	71.5	78
browsecomp	84.7	90.8
coding_arena_elo	1540	1662
critpt	16.9	29.1
DeepSWE 1.1	54	68.8
GDP (Surge AI)	81.6	85.5
gdpval	54.8	67.2
GDPval-AA v2	1618	1861
GPQA Diamond	91.1	93.7
HealthBench (raw)	59.2	67.1
HealthBench Professional (raw)	62.4	73.4
healthbench_professional	57.8	59.8
HLE	57.4	64.7
Legal Agent Benchmark	5.8	23
livebench_agentic_coding	59.4	65.2
livebench_coding	80.7	81.5
livebench_data_analysis	71.7	74.5
livebench_instruction_following	63.9	63.8
livebench_language	75	88.7
livebench_math	92.9	95.7
livebench_reasoning	88.7	91.2
MMMU-Pro	77.3	84.7
officeqa	73.3	78.1
officeqa_pro	59.4	66.9
OmniScience Accuracy	40	60.9
OmniScience Non-Hallucination	60.6	40.5
OSWorld-Verified	81.2	4
scicode	53.6	55.7
simplebench	60.6	80.6
simpleqa_verified	25	56.7
swe_bench_multimodal	28.1	59.4
SWE-bench Multilingual	78.3	89.5
SWE-bench Pro	63.2	79.2
SWE-bench Verified	85.2	96
TauBench V3 - Banking	37.3	44.7
Terminal-Bench 2.1	80.5	89.1
Toolathlon Verified Pass@3	84.3	87

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.