Which is better, Claude Opus 4.5 or Claude Sonnet 4.6?

Across 47 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 16 and Claude Sonnet 4.6 leads on 31 — Claude Sonnet 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 4.5 vs Claude Sonnet 4.6

AnthropicvsAnthropic47 shared benchmarks16–31 head-to-head

Benchmark	Claude Opus 4.5	Claude Sonnet 4.6
AA Agentic Index	59.6	61.6
AA Intelligence	41.9	48.4
AA-LCR	76	74
AA-Omniscience	14	12.2
arc_agi_1	82	86.5
ARC-AGI-2	37.6	60.4
arena_elo	1473	1472
arena_text_factuality	1458	1460
Artificial Analysis Coding Index	47.8	63
browsecomp	37	76.2
charxiv_rq	68.5	72.4
coding_arena_elo	1495	1523
critpt	4.6	3.1
finance_agent	55.2	63.3
frontiermath_tier_4	4.2	8.3
gdpval	47.3	54.8
GPQA Diamond	87	89.9
HLE	30.8	49
HLE (with tools)	43.4	46.8
IFBench	58	56.6
livebench	76	75.5
livebench_agentic_coding	39.7	42.6
livebench_coding	79.7	79.3
livebench_data_analysis	74.4	78
livebench_instruction_following	62.5	63.2
livebench_language	81.3	76.1
livebench_math	90.4	87
livebench_reasoning	80.1	84.8
MCP Atlas	69.8	69.5
mmmlu	90.8	89.3
MMMU-Pro	74	75.6
OmniScience Accuracy	46.6	40.9
OmniScience Non-Hallucination	39	51.6
OSWorld-Verified	66.3	78.5
scicode	50	47
simpleqa_verified	41.8	29
SWE-bench Pro	57.1	58.1
SWE-bench Verified	81.5	79.6
Tau2 retail	88.9	91.7
Terminal-Bench 2.0	59.3	59.1
Terminal-Bench Hard	47	59.1
toolathlon	43.5	49.4
vectara_answer_rate	98.7	99.9
vectara_avg_summary_length	114.5	114.7
vectara_factual_consistency	89.1	89.4
vectara_hallucination_rate ↓	10.9	10.6
τ²-Bench Telecom (AA run)	89.5	97.9

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.