Which is better, Claude Opus 4.7 or Claude Sonnet 4.5?

Across 44 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 41 and Claude Sonnet 4.5 leads on 1 — 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 4.5

AnthropicvsAnthropic44 shared benchmarks41–1 head-to-head

Benchmark	Claude Opus 4.7	Claude Sonnet 4.5
AA Agentic Index	64.6	50.6
AA Intelligence	55	37.4
AA-LCR	75.3	68.3
AA-Omniscience	27.3	-0.1
arc_agi_1	92	63.7
ARC-AGI-2	75.8	13.6
arena_elo	1494	1456
arena_text_factuality	1474	1448
Artificial Analysis Coding Index	73.6	52.1
browsecomp	79.8	24.1
charxiv_rq	82.1	67.2
coding_arena_elo	1557	1392
critpt	12	1.1
frontiermath_tier_4	22.9	4.2
gdpval	58.6	40.4
GPQA Diamond	94.2	83.4
HLE	54.7	19.8
HLE (with tools)	54.7	32
IFBench	58.6	57.3
MCP Atlas	79.1	59.5
mmmlu	91.5	89.1
MMMU-Pro	78.8	68.7
OmniScience Accuracy	48.9	32.9
OmniScience Non-Hallucination	57.7	50.9
OSWorld-Verified	82.8	61.4
scicode	54.5	45
simplebench	61.7	54.3
simpleqa_verified	50.6	23.6
SWE-bench Multilingual	80.5	68
SWE-bench Pro	64.3	43.6
SWE-bench Verified	87.6	82
TauBench V3 - Banking	34.6	24.5
Terminal-Bench 2.0	69.4	50
Terminal-Bench 2.1	83.1	55.8
Terminal-Bench Hard	54.5	35.6
toolathlon	59.3	41
Toolathlon Avg turns	25.9	32
Toolathlon Pass@∞	52.8	28.7
vectara_answer_rate	98	95.6
vectara_avg_summary_length	149.1	127.8
vectara_factual_consistency	88	88
vectara_hallucination_rate ↓	12	12
τ²-Bench Telecom (AA run)	88.6	78.1
τ³-Bench	28.9	19

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.