Which is better, Claude Opus 4.8 or Claude Sonnet 4.6?

Across 69 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 58 and Claude Sonnet 4.6 leads on 11 — Claude Opus 4.8 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 4.8 vs Claude Sonnet 4.6

AnthropicvsAnthropic69 shared benchmarks58–11 head-to-head

Benchmark	Claude Opus 4.8	Claude Sonnet 4.6
AA Agentic Index	49.4	61.6
AA Intelligence	57.3	48.4
AA-LCR	73	74
AA-Omniscience	28.8	12.2
AgentWorldBench - Android	61.5	58
AgentWorldBench - MCP	54.9	70
AgentWorldBench - OS	66.6	63.2
AgentWorldBench - Overall	56.6	56
AgentWorldBench - Search	35.1	28.8
AgentWorldBench - SWE	64.1	64.5
AgentWorldBench - Terminal	59.2	57
AgentWorldBench - Web	54.7	50.8
arc_agi_1	92.5	86.5
ARC-AGI-2	72.1	60.4
arena_elo	1482	1472
arena_text_factuality	1464	1460
arena_vision	1280	1278
Artificial Analysis Coding Index	74.3	63
browsecomp	88.5	76.2
ChartMuseum (with tools)	89.7	80.9
ChartMuseum (without tools)	75.8	59.3
charxiv_reasoning_with_tools	89.9	85.3
charxiv_rq	80.5	72.4
coding_arena_elo	1539	1523
critpt	20.9	3.1
DeepSWE 1.1	59	30
Finance Agent v2	53.9	51
finance_agent	53.9	63.3
frontiermath_tier_4	31.3	8.3
GDP (Surge AI)	84.8	78.6
gdpval	54.2	54.8
GDPval-AA (Elo)	1890	1676
GDPval-AA v2	1593	1381
GPQA Diamond	93.6	89.9
healthbench_professional	57.4	44.2
HLE	57.9	49
HLE (with tools)	57.9	46.8
IFBench	62.2	56.6
Legal Agent Benchmark	10	5.4
Legal Agent Benchmark, Full Public Set	9.6	8
livebench	77.2	75.5
livebench_agentic_coding	50.5	42.6
livebench_coding	81.8	79.3
livebench_data_analysis	66	78
livebench_instruction_following	72	63.2
livebench_language	79.7	76.1
livebench_math	94.3	87
livebench_reasoning	89.2	84.8
MCP Atlas	83.6	69.5
MMMU-Pro	78.9	75.6
officeqa	77.6	68.7
officeqa_pro	66.2	53.4
OmniScience Accuracy	48.8	40.9
OmniScience Non-Hallucination	60.7	51.6
OSWorld-Verified	83.4	78.5
scicode	53.5	47
simpleqa_verified	39.5	29
SWE-bench Pro	69.2	58.1
SWE-bench Verified	88.6	79.6
TauBench V3 - Banking	34.2	34.4
Terminal-Bench 2.0	74.6	59.1
Terminal-Bench 2.1	85	71.2
Terminal-Bench Hard	58.3	59.1
toolathlon	59.9	49.4
Toolathlon Avg turns	24.5	16.5
Toolathlon Pass@∞	48.1	38
usamo_2026	96.7	55
τ²-Bench Telecom (AA run)	94.4	97.9
τ³-Bench	27.6	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.