Which is better, Claude Opus 4.6 or Claude Sonnet 5?

Across 42 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 17 and Claude Sonnet 5 leads on 25 — Claude Sonnet 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 4.6 vs Claude Sonnet 5

AnthropicvsAnthropic42 shared benchmarks17–25 head-to-head

Benchmark	Claude Opus 4.6	Claude Sonnet 5
AA Agentic Index	67.6	49.7
AA Intelligence	44.9	55.3
AA-LCR	74.3	77
AA-Omniscience	13.7	16.4
arena_elo	1497	1463
arena_text_factuality	1487	1453
Artificial Analysis Coding Index	48.1	71.5
browsecomp	86.8	84.7
charxiv_reasoning_with_tools	84.7	88.3
charxiv_rq	69.1	77
coding_arena_elo	1545	1540
critpt	12.6	16.9
gdpval	55.9	54.8
GPQA Diamond	91.3	91.1
HLE	53.1	57.4
Legal Agent Benchmark	4.2	5.8
livebench_agentic_coding	49	59.4
livebench_coding	78.2	80.7
livebench_data_analysis	69.9	71.7
livebench_instruction_following	63.3	63.9
livebench_language	83.3	75
livebench_math	89.3	92.9
livebench_reasoning	88.7	88.7
MMMU-Pro	77.3	77.3
mrcr_v2_8needle_128k_average	84	81.5
officeqa	73.5	73.3
officeqa_pro	57.1	59.4
OmniScience Accuracy	47	40
OmniScience Non-Hallucination	37.2	60.6
OSWorld-Verified	72.7	81.2
scicode	52	53.6
simplebench	67.6	60.6
simpleqa_verified	46.5	25
swe_bench_multimodal	27.1	28.1
SWE-bench Multilingual	77.8	78.3
SWE-bench Pro	57.3	63.2
SWE-bench Verified	80.8	85.2
Terminal-Bench 2.0	65.4	80.4
toolathlon	56.8	54.3
Toolathlon Pass@∞	47.2	40.7
usamo_2026	66.2	79.5
τ³-Bench	72.4	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.