Which is better, Claude Opus 4.7 or Opus 5?

Across 44 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 10 and Opus 5 leads on 34 — 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 Opus 4.7 vs Opus 5

AnthropicvsAnthropic44 shared benchmarks10–34 head-to-head

Benchmark	Claude Opus 4.7	Opus 5
AA Agentic Index	64.6	59.2
AA Intelligence	55	63.1
AA-LCR	75.3	78.7
AA-Omniscience	27.3	37.1
arc_agi_1	92	97.5
arc_agi_3	0.2	30.2
ARC-AGI-2	75.8	90.4
arena_elo	1494	1487
arena_text_factuality	1474	1481
Artificial Analysis Coding Index	73.6	78
browsecomp	79.8	90.8
coding_arena_elo	1557	1662
critpt	12	29.1
Finance Agent v2	51.5	58.6
gdpval	58.6	67.2
GDPval-AA (Elo)	1753	1827
GPQA Diamond	94.2	93.7
HiL-Bench	41.7	57
HLE	54.7	64.7
HLE (with tools)	54.7	64.7
Legal Agent Benchmark	7.1	23
livebench_agentic_coding	50.7	65.2
livebench_coding	82.1	81.5
livebench_data_analysis	78.3	74.5
livebench_instruction_following	66.7	63.8
livebench_language	77.9	88.7
livebench_math	92.8	95.7
livebench_reasoning	87.2	91.2
MCP Atlas	79.1	85.8
MMMU-Pro	78.8	84.7
officeqa	86.3	78.1
officeqa_pro	80.6	66.9
OmniScience Accuracy	48.9	60.9
OmniScience Non-Hallucination	57.7	40.5
OSWorld-Verified	82.8	4
scicode	54.5	55.7
simplebench	61.7	80.6
simpleqa_verified	50.6	56.7
swe_bench_multimodal	34.5	59.4
SWE-bench Multilingual	80.5	89.5
SWE-bench Pro	64.3	79.2
SWE-bench Verified	87.6	96
TauBench V3 - Banking	34.6	44.7
Terminal-Bench 2.1	83.1	89.1

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.