Which is better, GPT-5.4 or Qwen3.5 397B A17B?

Across 49 shared benchmarks tracked on Vector Wire, GPT-5.4 leads on 45 and Qwen3.5 397B A17B leads on 4 — GPT-5.4 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.4 vs Qwen3.5 397B A17B

OpenAIvsAlibaba49 shared benchmarks45–4 head-to-head

Benchmark	GPT-5.4	Qwen3.5 397B A17B
AA Agentic Index	58.2	53.3
AA Intelligence	53.1	34.3
AA-LCR	77.7	72.7
AA-Omniscience	5.8	-30.8
AgentWorldBench - Android	60	54.9
AgentWorldBench - MCP	70.1	68.3
AgentWorldBench - OS	68.6	60.9
AgentWorldBench - Overall	58.3	54.7
AgentWorldBench - Search	37.3	30.8
AgentWorldBench - SWE	66.3	64.4
AgentWorldBench - Terminal	53.7	55.3
AgentWorldBench - Web	51.8	48.5
aime_2026	99.2	94.2
apexAgents	33.3	15.3
arena_elo	1477	1442
arena_text_factuality	1476	1445
arena_vision	1277	1249
Artificial Analysis Coding Index	71.1	48.2
browsecomp	82.7	69
charxiv_rq	82.8	80.8
coding_arena_elo	1457	1399
critpt	23.4	1.7
gdpval	50.1	35.8
GPQA Diamond	93	89.3
HLE	43.7	29
HLE (with tools)	52.1	48.3
hmmt_feb_2026	97.7	87.9
hmmt_nov_2025	95.8	92.7
IFBench	73.9	78.8
imo_answer_bench	91.4	80.9
itbenchSre	34.5	34.1
mcpmark	62.5	46.1
MMLU-Pro	87.5	87.8
MMMU-Pro	81.2	79
nl2repo	41.3	32.2
OmniScience Accuracy	50.9	30.8
OmniScience Non-Hallucination	17.4	17.3
scicode	56.6	42
simpleqa_verified	45.3	26
SWE-bench Multilingual	71.7	69.3
SWE-bench Pro	59.1	50.9
TauBench V3 - Banking	39.6	13.4
Terminal-Bench 2.0	75.1	52.5
Terminal-Bench 2.1	78.3	51.3
Terminal-Bench Hard	57.6	40.9
toolathlon	54.6	38.3
usamo_2026	95.2	36.3
τ²-Bench Telecom (AA run)	87.1	95.6
τ³-Bench	72.9	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.