Which is better, Claude Opus 4.6 or Claude Sonnet 4?

Across 45 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 39 and Claude Sonnet 4 leads on 6 — Claude Opus 4.6 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 4

AnthropicvsAnthropic45 shared benchmarks39–6 head-to-head

Benchmark	Claude Opus 4.6	Claude Sonnet 4
AA Agentic Index	67.6	39.2
AA Intelligence	44.9	29.8
AA-LCR	74.3	69.7
AA-Omniscience	13.7	0.2
AIME 2025	99.8	74
arc_agi_1	93	63.7
ARC-AGI-2	68.8	13.6
arena_vision	1300	1207
Artificial Analysis Coding Index	48.1	37.6
browsecomp	86.8	12.2
critpt	12.6	1.1
Fortress	20.5	24.4
frontiermath_tier_4	22.9	0
gdpval	55.9	31.2
GPQA Diamond	91.3	78
HLE	53.1	10.7
HLE (with tools)	62.7	20.3
IFBench	62.5	55
livebench	76.3	74.8
LiveCodeBench	88.8	68.5
LiveCodeBench v6	88.8	48.5
MMLU-Pro	89.1	84
mmmlu	91.1	86.5
MMMU-Pro	77.3	62.4
multichallenge	56	57.1
MultiNRC	57.1	18.4
OmniScience Accuracy	47	22.7
OmniScience Non-Hallucination	37.2	70.9
OSWorld-Verified	72.7	42.2
scicode	52	40
SEAL VISTA	46.1	45.5
simplebench	67.6	45.5
swe_bench_bash	75.6	64.9
SWE-bench Multilingual	77.8	56.9
SWE-bench Pro	57.3	42.7
SWE-bench Verified	80.8	80.2
Tau2 retail	91.9	75
Terminal-Bench Hard	65.4	31.1
vectara_answer_rate	99.8	98.6
vectara_avg_summary_length	137.6	145.8
vectara_factual_consistency	87.8	89.7
vectara_hallucination_rate ↓	12.2	10.3
VisualToolBench	27.5	4.5
τ²-Bench Telecom (AA run)	99.3	64.6
τ³-Bench	72.4	13.8

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.