Which is better, Claude Sonnet 4.6 or Qwen3.5 397B A17B?

Across 44 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 39 and Qwen3.5 397B A17B leads on 5 — 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 Sonnet 4.6 vs Qwen3.5 397B A17B

AnthropicvsAlibaba44 shared benchmarks39–5 head-to-head

Benchmark	Claude Sonnet 4.6	Qwen3.5 397B A17B
AA Agentic Index	61.6	53.3
AA Intelligence	48.4	34.3
AA-LCR	74	72.7
AA-Omniscience	12.2	-30.8
AgentWorldBench - Android	58	54.9
AgentWorldBench - MCP	70	68.3
AgentWorldBench - OS	63.2	60.9
AgentWorldBench - Overall	56	54.7
AgentWorldBench - Search	28.8	30.8
AgentWorldBench - SWE	64.5	64.4
AgentWorldBench - Terminal	57	55.3
AgentWorldBench - Web	50.8	48.5
apexAgents	28	15.3
arena_elo	1472	1442
arena_text_factuality	1460	1445
arena_vision	1278	1249
Artificial Analysis Coding Index	63	48.2
browsecomp	76.2	69
charxiv_rq	72.4	80.8
coding_arena_elo	1523	1399
critpt	3.1	1.7
gdpval	54.8	35.8
GDPval-AA v2	1381	962
GPQA Diamond	89.9	89.3
HLE	49	29
HLE (with tools)	46.8	48.3
IFBench	56.6	78.8
itbenchSre	39.8	34.1
mmmlu	89.3	88.5
MMMU-Pro	75.6	79
OmniScience Accuracy	40.9	30.8
OmniScience Non-Hallucination	51.6	17.3
scicode	47	42
simpleqa_verified	29	26
SWE-bench Pro	58.1	50.9
SWE-bench Verified	79.6	76.4
TauBench V3 - Banking	34.4	13.4
Terminal-Bench 2.0	59.1	52.5
Terminal-Bench 2.1	71.2	51.3
Terminal-Bench Hard	59.1	40.9
toolathlon	49.4	38.3
usamo_2026	55	36.3
τ²-Bench Telecom (AA run)	97.9	95.6
τ³-Bench	30.5	13.4

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.