Which is better, Gemma 4 31B or Qwen3.5 397B A17B?

Across 59 shared benchmarks tracked on Vector Wire, Gemma 4 31B leads on 9 and Qwen3.5 397B A17B leads on 50 — 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 —

Gemma 4 31B vs Qwen3.5 397B A17B

GooglevsAlibaba59 shared benchmarks9–50 head-to-head

Benchmark	Gemma 4 31B	Qwen3.5 397B A17B
AA Agentic Index	14.4	53.3
AA Intelligence	29.7	34.3
AA-LCR	68.3	72.7
AA-Omniscience	-47.9	-30.8
aime_2026	89.2	94.2
arena_elo	1451	1442
arena_text_factuality	1449	1445
arena_vision	1252	1249
Artificial Analysis Coding Index	43.4	48.2
C-Eval	82.6	93
CC-OCR	75.7	82
charxiv_rq	67.9	80.8
Claw Eval (pass@3)	25	48.1
Claw-Eval Avg	48.5	70.7
coding_arena_elo	1364	1399
critpt	1.4	1.7
DeepPlanning	24	34.3
DynaMath	79.5	86.3
ERQA	57.5	67.5
gdpval	15.5	35.8
GPQA Diamond	85.7	89.3
HLE	26.5	29
HLE (with tools)	26.5	48.3
hmmt_feb_2025	88.7	94.8
hmmt_feb_2026	77.2	87.9
hmmt_nov_2025	87.5	92.7
IFBench	75.6	78.8
imo_answer_bench	74.5	80.9
itbenchSre	37.3	34.1
LiveCodeBench v6	80	83.6
mcpmark	18.1	46.1
mmlu_redux	93.7	94.9
MMLU-Pro	85.2	87.8
mmmlu	88.4	88.5
MMMU	80.4	85
MMMU-Pro	76.9	79
MMStar	77.3	83.8
nl2repo	15.5	32.2
OmniScience Accuracy	20	30.8
OmniScience Non-Hallucination	18.1	17.3
QwenClawBench	41.7	51.8
QwenWebBench	1197	1186
RealWorldQA	72.3	83.9
scicode	43.4	42
simpleqa_verified	9.6	26
SimpleVQA	52.9	67.1
SkillsBench Avg5	23.6	30
supergpqa	65.7	70.4
SWE-bench Multilingual	51.7	69.3
SWE-bench Pro	35.7	50.9
SWE-bench Verified	52	76.4
TauBench V3 - Banking	14.8	13.4
Terminal-Bench 2.0	42.9	52.5
Terminal-Bench 2.1	43.4	51.3
Terminal-Bench Hard	36.4	40.9
VITA-Bench	43	49.7
WideSearch	35.2	74
τ²-Bench Telecom (AA run)	65.5	95.6
τ³-Bench	67.5	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.