Which is better, Claude Opus 4.8 or Claude Sonnet 5?

Across 53 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 39 and Claude Sonnet 5 leads on 14 — 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 5

AnthropicvsAnthropic53 shared benchmarks39–14 head-to-head

Benchmark	Claude Opus 4.8	Claude Sonnet 5
AA Agentic Index	49.4	49.7
AA Intelligence	57.3	55.3
AA-LCR	73	77
AA-Omniscience	28.8	16.4
arena_elo	1482	1463
arena_text_factuality	1464	1453
Artificial Analysis Coding Index	74.3	71.5
browsecomp	88.5	84.7
ChartMuseum (with tools)	89.7	86.7
ChartMuseum (without tools)	75.8	70.1
charxiv_reasoning_with_tools	89.9	88.3
charxiv_rq	80.5	77
coding_arena_elo	1539	1540
critpt	20.9	16.9
DeepSWE 1.1	59	54
GDP (Surge AI)	84.8	81.6
gdpval	54.2	54.8
GDPval-AA v2	1593	1618
GPQA Diamond	93.6	91.1
HealthBench (raw)	58.8	59.2
HealthBench Professional (raw)	60.3	62.4
healthbench_professional	57.4	57.8
HLE	57.9	57.4
Legal Agent Benchmark	10	5.8
Legal Agent Benchmark, Full Public Set	9.6	8.9
livebench_agentic_coding	50.5	59.4
livebench_coding	81.8	80.7
livebench_data_analysis	66	71.7
livebench_instruction_following	72	63.9
livebench_language	79.7	75
livebench_math	94.3	92.9
livebench_reasoning	89.2	88.7
MMMU-Pro	78.9	77.3
officeqa	77.6	73.3
officeqa_pro	66.2	59.4
OmniScience Accuracy	48.8	40
OmniScience Non-Hallucination	60.7	60.6
OSWorld-Verified	83.4	81.2
scicode	53.5	53.6
simplebench	64.8	60.6
simpleqa_verified	39.5	25
swe_bench_multimodal	38.4	28.1
SWE-bench Multilingual	84.4	78.3
SWE-bench Pro	69.2	63.2
SWE-bench Verified	88.6	85.2
TauBench V3 - Banking	34.2	37.3
Terminal-Bench 2.0	74.6	80.4
Terminal-Bench 2.1	85	80.5
toolathlon	59.9	54.3
Toolathlon Pass@∞	48.1	40.7
Toolathlon Verified Pass@3	88	84.3
usamo_2026	96.7	79.5
τ³-Bench	27.6	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.