Which is better, Claude Sonnet 4.5 or Qwen3.5 35B A3B?

Across 55 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 23 and Qwen3.5 35B A3B leads on 31 — Qwen3.5 35B A3B 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 35B A3B

AnthropicvsAlibaba55 shared benchmarks23–31 head-to-head

Benchmark	Claude Sonnet 4.5	Qwen3.5 35B A3B
AA Agentic Index	50.6	44.1
AA Intelligence	37.4	29.9
AA-LCR	68.3	68.3
AA-Omniscience	-0.1	-48.1
AI2D	87	92.6
Artificial Analysis Coding Index	52.1	37
browsecomp	24.1	61
browsecomp_zh	42.4	69.5
CC-OCR	68.1	80.7
charxiv_rq	67.2	77.5
coding_arena_elo	1392	1250
critpt	1.1	0.9
gdpval	40.4	20.3
GPQA (unspecified)	83.4	84.2
GPQA Diamond	83.4	84.5
HallusionBench	59.9	67.9
HLE	19.8	47.4
HMMT 2025	74.6	89.2
hmmt_feb_2025	79.2	89
hmmt_nov_2025	81.7	89.2
IFBench	57.3	72.5
imo_answer_bench	65.9	76.8
LiveCodeBench v6	64	74.6
longbench_v2	61.8	59
MathVista	79.8	86.2
MCP Atlas	59.5	62.4
MMBench EN-DEV-v1.1	88.3	91.5
mmlu_redux	95.6	93.3
MMLU-Pro	88.2	85.3
mmmlu	89.1	85.2
MMMU	79.6	81.4
MMMU-Pro	68.7	75.1
multichallenge	55.3	60
OmniDocBench 1.5	85.8	89.3
OmniScience Accuracy	32.9	20.1
OmniScience Non-Hallucination	50.9	14.6
OSWorld-Verified	61.4	54.5
RealWorldQA	70.3	84.1
scicode	45	37.7
Seal-0	53.4	41.4
SimpleVQA	57.6	58.3
SWE-bench Multilingual	68	60.3
SWE-bench Pro	43.6	44.6
SWE-bench Verified	82	70
TauBench V3 - Banking	24.5	4.9
Terminal-Bench 2.0	50	40.5
Terminal-Bench 2.1	55.8	40.8
Terminal-Bench Hard	35.6	26.5
vectara_answer_rate	95.6	99.8
vectara_avg_summary_length	127.8	94.9
vectara_factual_consistency	88	89.5
vectara_hallucination_rate ↓	12	10.5
ZEROBench_sub	26.3	34.1
τ²-Bench Telecom (AA run)	78.1	89.2
τ³-Bench	19	68.9

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.