Which is better, Gemma 4 31B or Qwen3.6 27B?

Across 54 shared benchmarks tracked on Vector Wire, Gemma 4 31B leads on 10 and Qwen3.6 27B leads on 44 — Qwen3.6 27B 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 Qwen3.6 27B

GooglevsAlibaba54 shared benchmarks10–44 head-to-head

Benchmark	Gemma 4 31B	Qwen3.6 27B
AA Agentic Index	14.4	27.5
AA Intelligence	29.7	37.7
AA-LCR	68.3	73.3
AA-Omniscience	-47.9	-20
aime_2026	89.2	94.1
Artificial Analysis Coding Index	43.4	53.7
C-Eval	82.6	91.4
CC-OCR	75.7	81.2
charxiv_rq	67.9	78.4
Claw Eval (pass@3)	25	60.6
Claw-Eval Avg	48.5	72.4
critpt	1.4	1.1
DynaMath	79.5	85.6
ERQA	57.5	62.5
gdpval	15.5	32
GPQA Diamond	85.7	87.8
HLE	26.5	24
hmmt_feb_2025	88.7	93.8
hmmt_feb_2026	77.2	84.3
hmmt_nov_2025	87.5	90.7
IFBench	75.6	69.1
imo_answer_bench	74.5	80.8
LiveCodeBench v6	80	83.9
mathvision	85.6	85.1
MathVista	79.3	87.4
MMBench EN-DEV-v1.1	90.9	92.3
mmlu_redux	93.7	93.5
MMLU-Pro	85.2	86.2
MMMU	80.4	82.9
MMMU-Pro	76.9	75.8
MMStar	77.3	81.4
nl2repo	15.5	36.2
OCRBench	86.1	89.4
OmniDocBench 1.5	80.1	89.4
OmniScience Accuracy	20	19.6
OmniScience Non-Hallucination	18.1	50.7
QwenClawBench	41.7	53.4
QwenWebBench	1197	1487
RealWorldQA	72.3	84.1
RefSpatial-Bench	4.7	70
scicode	43.4	39.8
SimpleVQA	52.9	56.1
SkillsBench Avg5	23.6	48.2
supergpqa	65.7	66
SWE-bench Multilingual	51.7	71.3
SWE-bench Pro	35.7	53.5
SWE-bench Verified	52	77.2
TauBench V3 - Banking	14.8	16.7
Terminal-Bench 2.0	42.9	59.3
Terminal-Bench 2.1	43.4	60.7
Terminal-Bench Hard	36.4	34.8
VlmsAreBlind	87.2	97
τ²-Bench Telecom (AA run)	65.5	94.2
τ³-Bench	67.5	15.3

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.