Which is better, Claude Opus 4.7 or Claude Sonnet 5?

Across 44 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 31 and Claude Sonnet 5 leads on 13 — Claude Opus 4.7 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 4.7 vs Claude Sonnet 5

AnthropicvsAnthropic44 shared benchmarks31–13 head-to-head

Benchmark	Claude Opus 4.7	Claude Sonnet 5
AA Agentic Index	64.6	49.7
AA Intelligence	55	55.3
AA-LCR	75.3	77
AA-Omniscience	27.3	16.4
arena_elo	1494	1463
arena_text_factuality	1474	1453
Artificial Analysis Coding Index	73.6	71.5
browsecomp	79.8	84.7
charxiv_reasoning_with_tools	91	88.3
charxiv_rq	82.1	77
coding_arena_elo	1557	1540
critpt	12	16.9
gdpval	58.6	54.8
GPQA Diamond	94.2	91.1
HLE	54.7	57.4
Legal Agent Benchmark	7.1	5.8
livebench_agentic_coding	50.7	59.4
livebench_coding	82.1	80.7
livebench_data_analysis	78.3	71.7
livebench_instruction_following	66.7	63.9
livebench_language	77.9	75
livebench_math	92.8	92.9
livebench_reasoning	87.2	88.7
MMMU-Pro	78.8	77.3
mrcr_v2_8needle_128k_average	59.3	81.5
officeqa	86.3	73.3
officeqa_pro	80.6	59.4
OmniScience Accuracy	48.9	40
OmniScience Non-Hallucination	57.7	60.6
OSWorld-Verified	82.8	81.2
scicode	54.5	53.6
simplebench	61.7	60.6
simpleqa_verified	50.6	25
swe_bench_multimodal	34.5	28.1
SWE-bench Multilingual	80.5	78.3
SWE-bench Pro	64.3	63.2
SWE-bench Verified	87.6	85.2
TauBench V3 - Banking	34.6	37.3
Terminal-Bench 2.0	69.4	80.4
Terminal-Bench 2.1	83.1	80.5
toolathlon	59.3	54.3
Toolathlon Pass@∞	52.8	40.7
usamo_2026	69.3	79.5
τ³-Bench	28.9	28.2

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.