Which is better, Claude Sonnet 4.5 or Qwen3.5 27B?

Across 52 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 22 and Qwen3.5 27B leads on 30 — Qwen3.5 27B 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 27B

AnthropicvsAlibaba52 shared benchmarks22–30 head-to-head

Benchmark	Claude Sonnet 4.5	Qwen3.5 27B
AA Agentic Index	50.6	54.6
AA Intelligence	37.4	34.6
AA-LCR	68.3	72.3
AA-Omniscience	-0.1	-44
AI2D	87	92.9
Artificial Analysis Coding Index	52.1	34.9
browsecomp	24.1	61
browsecomp_zh	42.4	62.1
CC-OCR	68.1	81
charxiv_rq	67.2	79.5
coding_arena_elo	1392	1357
critpt	1.1	0.9
gdpval	40.4	33
GPQA Diamond	83.4	85.8
HallusionBench	59.9	70
HLE	19.8	48.5
HMMT 2025	74.6	89.8
hmmt_feb_2025	79.2	92
hmmt_nov_2025	81.7	89.8
IFBench	57.3	76.5
imo_answer_bench	65.9	79.9
LiveCodeBench v6	64	80.7
longbench_v2	61.8	60.6
MathVista	79.8	87.8
MCP Atlas	59.5	68.4
MMBench EN-DEV-v1.1	88.3	92.6
mmlu_redux	95.6	93.2
MMLU-Pro	88.2	86.1
mmmlu	89.1	85.9
MMMU	79.6	82.3
MMMU-Pro	68.7	75
multichallenge	55.3	60.8
OmniDocBench 1.5	85.8	88.9
OmniScience Accuracy	32.9	20.7
OmniScience Non-Hallucination	50.9	24.9
OSWorld-Verified	61.4	56.2
RealWorldQA	70.3	83.7
scicode	45	39.5
Seal-0	53.4	47.2
SimpleVQA	57.6	56
SWE-bench Multilingual	68	69.3
SWE-bench Pro	43.6	51.2
SWE-bench Verified	82	75
Terminal-Bench 2.0	50	41.6
Terminal-Bench Hard	35.6	32.6
vectara_answer_rate	95.6	99.8
vectara_avg_summary_length	127.8	94.4
vectara_factual_consistency	88	87.9
vectara_hallucination_rate ↓	12	12.1
ZEROBench_sub	26.3	36.2
τ²-Bench Telecom (AA run)	78.1	93.9
τ³-Bench	19	68.4

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.