Which is better, Claude Sonnet 4.6 or Claude Sonnet 5?

Across 50 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 10 and Claude Sonnet 5 leads on 40 — Claude Sonnet 5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Sonnet 4.6 vs Claude Sonnet 5

AnthropicvsAnthropic50 shared benchmarks10–40 head-to-head

Benchmark	Claude Sonnet 4.6	Claude Sonnet 5
AA Agentic Index	61.6	49.7
AA Intelligence	48.4	55.3
AA-LCR	74	77
AA-Omniscience	12.2	16.4
arena_elo	1472	1463
arena_text_factuality	1460	1453
Artificial Analysis Coding Index	63	71.5
BenchCAD Vision2Code (with tools)	0.3	0.4
BenchCAD Vision2Code (without tools)	0.3	0.3
browsecomp	76.2	84.7
ChartMuseum (with tools)	80.9	86.7
ChartMuseum (without tools)	59.3	70.1
charxiv_reasoning_with_tools	85.3	88.3
charxiv_rq	72.4	77
coding_arena_elo	1523	1540
critpt	3.1	16.9
DeepSWE 1.1	30	54
GDP (Surge AI)	78.6	81.6
gdpval	54.8	54.8
GDPval-AA v2	1381	1618
GPQA Diamond	89.9	91.1
healthbench_professional	44.2	57.8
HLE	49	57.4
Legal Agent Benchmark	5.4	5.8
Legal Agent Benchmark, Full Public Set	8	8.9
livebench_agentic_coding	42.6	59.4
livebench_coding	79.3	80.7
livebench_data_analysis	78	71.7
livebench_instruction_following	63.2	63.9
livebench_language	76.1	75
livebench_math	87	92.9
livebench_reasoning	84.8	88.7
MMMU-Pro	75.6	77.3
mrcr_v2_8needle_128k_average	84.9	81.5
officeqa	68.7	73.3
officeqa_pro	53.4	59.4
OmniScience Accuracy	40.9	40
OmniScience Non-Hallucination	51.6	60.6
OSWorld-Verified	78.5	81.2
scicode	47	53.6
simpleqa_verified	29	25
SWE-bench Pro	58.1	63.2
SWE-bench Verified	79.6	85.2
TauBench V3 - Banking	34.4	37.3
Terminal-Bench 2.0	59.1	80.4
Terminal-Bench 2.1	71.2	80.5
toolathlon	49.4	54.3
Toolathlon Pass@∞	38	40.7
usamo_2026	55	79.5
τ³-Bench	30.5	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.