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

Across 67 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 51 and Claude Sonnet 4.6 leads on 16 — 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 4.6

AnthropicvsAnthropic67 shared benchmarks51–16 head-to-head

Benchmark	Claude Opus 4.6	Claude Sonnet 4.6
AA Agentic Index	67.6	61.6
AA Intelligence	44.9	48.4
AA-LCR	74.3	74
AA-Omniscience	13.7	12.2
AgentWorldBench - Android	61.7	58
AgentWorldBench - MCP	69.9	70
AgentWorldBench - OS	70.2	63.2
AgentWorldBench - Overall	57.8	56
AgentWorldBench - Search	29.3	28.8
AgentWorldBench - SWE	64.5	64.5
AgentWorldBench - Terminal	57.5	57
AgentWorldBench - Web	51.4	50.8
apexAgents	33	28
arc_agi_1	93	86.5
ARC-AGI-2	68.8	60.4
arena_elo	1497	1472
arena_text_factuality	1487	1460
arena_vision	1300	1278
Artificial Analysis Coding Index	48.1	63
browsecomp	86.8	76.2
charxiv_reasoning_with_tools	84.7	85.3
charxiv_rq	69.1	72.4
coding_arena_elo	1545	1523
critpt	12.6	3.1
finance_agent	76.7	63.3
frontiermath_tier_4	22.9	8.3
gdpval	55.9	54.8
GDPval-AA (Elo)	1619	1676
GPQA Diamond	91.3	89.9
HLE	53.1	49
HLE (with tools)	62.7	46.8
IFBench	62.5	56.6
Legal Agent Benchmark	4.2	5.4
livebench	76.3	75.5
livebench_agentic_coding	49	42.6
livebench_coding	78.2	79.3
livebench_data_analysis	69.9	78
livebench_instruction_following	63.3	63.2
livebench_language	83.3	76.1
livebench_math	89.3	87
livebench_reasoning	88.7	84.8
MCP Atlas	76.8	69.5
mmmlu	91.1	89.3
MMMU-Pro	77.3	75.6
mrcr_v2_8needle_128k_average	84	84.9
officeqa	73.5	68.7
officeqa_pro	57.1	53.4
OmniScience Accuracy	47	40.9
OmniScience Non-Hallucination	37.2	51.6
OSWorld-Verified	72.7	78.5
scicode	52	47
simpleqa_verified	46.5	29
SWE-bench Pro	57.3	58.1
SWE-bench Verified	80.8	79.6
Tau2 retail	91.9	91.7
Terminal-Bench 2.0	65.4	59.1
Terminal-Bench Hard	65.4	59.1
toolathlon	56.8	49.4
Toolathlon Avg turns	16.9	16.5
Toolathlon Pass@∞	47.2	38
usamo_2026	66.2	55
vectara_answer_rate	99.8	99.9
vectara_avg_summary_length	137.6	114.7
vectara_factual_consistency	87.8	89.4
vectara_hallucination_rate ↓	12.2	10.6
τ²-Bench Telecom (AA run)	99.3	97.9
τ³-Bench	72.4	30.5

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.