Which is better, Claude Sonnet 4.6 or Opus 5?

Across 43 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 4 and Opus 5 leads on 39 — 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 4.6 vs Opus 5

AnthropicvsAnthropic43 shared benchmarks4–39 head-to-head

Benchmark	Claude Sonnet 4.6	Opus 5
AA Agentic Index	61.6	59.2
AA Intelligence	48.4	63.1
AA-LCR	74	78.7
AA-Omniscience	12.2	37.1
arc_agi_1	86.5	97.5
ARC-AGI-2	60.4	90.4
arena_elo	1472	1487
arena_text_factuality	1460	1481
Artificial Analysis Coding Index	63	78
browsecomp	76.2	90.8
coding_arena_elo	1523	1662
critpt	3.1	29.1
DeepSWE 1.1	30	68.8
Finance Agent v2	51	58.6
GDP (Surge AI)	78.6	85.5
gdpval	54.8	67.2
GDPval-AA (Elo)	1676	1827
GDPval-AA v2	1381	1861
GPQA Diamond	89.9	93.7
healthbench_professional	44.2	59.8
HLE	49	64.7
HLE (with tools)	46.8	64.7
Legal Agent Benchmark	5.4	23
livebench_agentic_coding	42.6	65.2
livebench_coding	79.3	81.5
livebench_data_analysis	78	74.5
livebench_instruction_following	63.2	63.8
livebench_language	76.1	88.7
livebench_math	87	95.7
livebench_reasoning	84.8	91.2
MCP Atlas	69.5	85.8
MMMU-Pro	75.6	84.7
officeqa	68.7	78.1
officeqa_pro	53.4	66.9
OmniScience Accuracy	40.9	60.9
OmniScience Non-Hallucination	51.6	40.5
OSWorld-Verified	78.5	4
scicode	47	55.7
simpleqa_verified	29	56.7
SWE-bench Pro	58.1	79.2
SWE-bench Verified	79.6	96
TauBench V3 - Banking	34.4	44.7
Terminal-Bench 2.1	71.2	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.