Which is better, GPT-5.4 or Qwen3.7 Max?

Across 44 shared benchmarks tracked on Vector Wire, GPT-5.4 leads on 29 and Qwen3.7 Max leads on 15 — 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 Qwen3.7 Max

OpenAIvsAlibaba44 shared benchmarks29–15 head-to-head

Benchmark	GPT-5.4	Qwen3.7 Max
AA Agentic Index	58.2	30.9
AA Intelligence	53.1	46.7
AA-LCR	77.7	74.7
AA-Omniscience	5.8	13.5
aime_2026	99.2	97
arena_elo	1477	1474
arena_text_factuality	1476	1479
Artificial Analysis Coding Index	71.1	66
coding_arena_elo	1457	1517
critpt	23.4	13.4
DeepSWE 1.1	52	21.6
gdpval	50.1	38.5
GPQA Diamond	93	92.4
HLE	43.7	41.4
HLE (with tools)	52.1	53.5
hmmt_feb_2026	97.7	97.1
hmmt_nov_2025	95.8	95
IFBench	73.9	80.5
imo_answer_bench	91.4	90
itbenchSre	34.5	42.5
livebench	80.3	74.3
livebench_agentic_coding	53.8	43.6
livebench_coding	77.5	74.2
livebench_data_analysis	79.3	71.8
livebench_instruction_following	70.2	74
livebench_language	82.6	79.7
livebench_math	94.2	85.3
livebench_reasoning	88.1	83.3
MCP Atlas	70.6	76.4
mcpmark	62.5	60.8
MMLU-Pro	87.5	89.6
nl2repo	41.3	47.2
OmniScience Accuracy	50.9	31.1
OmniScience Non-Hallucination	17.4	74.4
scicode	56.6	53.5
simpleqa_verified	45.3	58.5
SWE-bench Multilingual	71.7	78.3
SWE-bench Pro	59.1	60.6
TauBench V3 - Banking	39.6	11.8
Terminal-Bench 2.0	75.1	69.7
Terminal-Bench 2.1	78.3	75
Terminal-Bench Hard	57.6	50.8
τ²-Bench Telecom (AA run)	87.1	94.7
τ³-Bench	72.9	10.9

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.