Which is better, Gemma 4 31B or Qwen3.6 35B A3B?

Across 58 shared benchmarks tracked on Vector Wire, Gemma 4 31B leads on 15 and Qwen3.6 35B A3B leads on 43 — Qwen3.6 35B A3B 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 35B A3B

GooglevsAlibaba58 shared benchmarks15–43 head-to-head

Benchmark	Gemma 4 31B	Qwen3.6 35B A3B
AA Agentic Index	14.4	21.6
AA Intelligence	29.7	32.1
AA-LCR	68.3	66.7
AA-Omniscience	-47.9	-22.2
AI2D	89	92.7
aime_2026	89.2	92.7
Artificial Analysis Coding Index	43.4	41.9
C-Eval	82.6	90
CC-OCR	75.7	81.9
charxiv_rq	67.9	78
Claw Eval (pass@3)	25	50
Claw-Eval Avg	48.5	68.7
critpt	1.4	0.3
DeepPlanning	24	25.9
gdpval	15.5	27.8
GPQA (unspecified)	84.3	86
GPQA Diamond	85.7	86
HallusionBench	67.4	69.8
HLE	26.5	22.2
hmmt_feb_2025	88.7	90.7
hmmt_feb_2026	77.2	83.6
hmmt_nov_2025	87.5	89.1
IFBench	75.6	64.4
imo_answer_bench	74.5	78.9
LiveCodeBench v6	80	80.4
MathVista	79.3	86.4
MCP Atlas	57.2	62.8
mcpmark	18.1	37
MMBench EN-DEV-v1.1	90.9	92.8
mmlu_redux	93.7	93.3
MMLU-Pro	85.2	85.6
MMMU	80.4	81.7
MMMU-Pro	76.9	75.3
nl2repo	15.5	29.4
OmniDocBench 1.5	80.1	89.9
OmniScience Accuracy	20	18.8
OmniScience Non-Hallucination	18.1	49.5
QwenClawBench	41.7	52.6
QwenWebBench	1197	1397
RealWorldQA	72.3	85.3
RefSpatial-Bench	4.7	0.6
scicode	43.4	35.8
SimpleVQA	52.9	58.9
SkillsBench Avg5	23.6	28.7
supergpqa	65.7	64.7
SWE-bench Multilingual	51.7	67.2
SWE-bench Pro	35.7	49.5
SWE-bench Verified	52	73.4
TauBench V3 - Banking	14.8	9.3
Terminal-Bench 2.0	42.9	51.5
Terminal-Bench 2.1	43.4	44.9
Terminal-Bench Hard	36.4	34.8
tool_decathlon	21.2	26.9
VITA-Bench	43	35.6
WideSearch	35.2	60.1
ZEROBench_sub	26	34.4
τ²-Bench Telecom (AA run)	65.5	95.3
τ³-Bench	67.5	67.2

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.