Which is better, Claude Sonnet 4.5 or Qwen3.5 397B A17B?

Across 48 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 18 and Qwen3.5 397B A17B leads on 30 — Qwen3.5 397B A17B 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 4.5 vs Qwen3.5 397B A17B

AnthropicvsAlibaba48 shared benchmarks18–30 head-to-head

Benchmark	Claude Sonnet 4.5	Qwen3.5 397B A17B
AA Agentic Index	50.6	53.3
AA Intelligence	37.4	34.3
AA-LCR	68.3	72.7
AA-Omniscience	-0.1	-30.8
arena_elo	1456	1442
arena_text_factuality	1448	1445
Artificial Analysis Coding Index	52.1	48.2
browsecomp	24.1	69
BrowseComp (context management)	26.1	78.6
browsecomp_zh	42.4	70.3
CC-OCR	68.1	82
charxiv_rq	67.2	80.8
coding_arena_elo	1392	1399
critpt	1.1	1.7
gdpval	40.4	35.8
GPQA Diamond	83.4	89.3
HLE	19.8	29
HLE (with tools)	32	48.3
HMMT 2025	74.6	92.7
hmmt_feb_2025	79.2	94.8
hmmt_nov_2025	81.7	92.7
IFBench	57.3	78.8
imo_answer_bench	65.9	80.9
LiveCodeBench v6	64	83.6
longbench_v2	61.8	63.2
mmlu_redux	95.6	94.9
MMLU-Pro	88.2	87.8
mmmlu	89.1	88.5
MMMU	79.6	85
MMMU-Pro	68.7	79
multichallenge	55.3	67.6
OmniScience Accuracy	32.9	30.8
OmniScience Non-Hallucination	50.9	17.3
RealWorldQA	70.3	83.9
scicode	45	42
Seal-0	53.4	46.9
simpleqa_verified	23.6	26
SimpleVQA	57.6	67.1
SWE-bench Multilingual	68	69.3
SWE-bench Pro	43.6	50.9
SWE-bench Verified	82	76.4
TauBench V3 - Banking	24.5	13.4
Terminal-Bench 2.0	50	52.5
Terminal-Bench 2.1	55.8	51.3
Terminal-Bench Hard	35.6	40.9
toolathlon	41	38.3
τ²-Bench Telecom (AA run)	78.1	95.6
τ³-Bench	19	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.