Which is better, Gemma 4 26B A4B or Qwen3.5 35B A3B?

Across 78 shared benchmarks tracked on Vector Wire, Gemma 4 26B A4B leads on 17 and Qwen3.5 35B A3B leads on 61 — Qwen3.5 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 26B A4B vs Qwen3.5 35B A3B

GooglevsAlibaba78 shared benchmarks17–61 head-to-head

Benchmark	Gemma 4 26B A4B	Qwen3.5 35B A3B
AA Agentic Index	32.1	44.1
AA Intelligence	26.1	29.9
AA-LCR	61.7	68.3
AA-Omniscience	-50.8	-48.1
AGIEval-En	55.3	70.5
AI2D	88.3	92.6
aime_2026	88.3	93.3
arc_challenge	92.8	95.4
Artificial Analysis Coding Index	39.3	37
browsecomp	26.3	61
C-Eval	82.5	90.2
CC-OCR	74.5	80.7
charxiv_rq	69	77.5
Claw Eval (pass@3)	28	51
Claw-Eval Avg	58.8	65.4
coding_arena_elo	1361	1250
critpt	0	0.9
DeepPlanning	16.2	22.8
gdpval	25.7	20.3
Global-MMLU-Lite (5-shot) Average	74.8	80.9
GPQA (unspecified)	82.3	84.2
GPQA Diamond	82.3	84.5
GSM8K	77	90.1
HallusionBench	66.1	67.9
hellaswag	85.3	85.6
HLE	19.3	47.4
hmmt_feb_2025	91.7	89
hmmt_feb_2026	79	81.8
hmmt_nov_2025	87.5	89.2
humaneval	50	66.5
IFBench	72.4	72.5
ifeval	96.4	91.9
imo_answer_bench	74.3	76.8
itbenchSre	23.6	21.5
LiveCodeBench v6	77.1	74.6
mathvision	82.4	83.9
MathVista	79.4	86.2
MBPP	68.4	70.8
MCP Atlas	50	62.4
mcpmark	14.2	27
Minerva Math	43.7	59.7
MMBench EN-DEV-v1.1	89	91.5
mmlu	77.8	81.1
mmlu_redux	92.7	93.3
MMLU-Pro	85.2	85.3
mmmlu	86.3	85.2
MMMU	78.4	81.4
MMMU-Pro	73.8	75.1
nl2repo	11.6	20.5
OmniDocBench 1.5	74.4	89.3
OmniScience Accuracy	19.1	20.1
OmniScience Non-Hallucination	13.6	14.6
OpenBookQA	48.8	44.2
PIQA (acc-norm)	83.8	82.5
PIQA (acc)	82.2	82.3
QwenClawBench	38.7	47.7
QwenWebBench	1178	978
RealWorldQA	72.2	84.1
RULER 1M	72.9	56.4
RULER 256K	85.7	82.4
scicode	40.3	37.7
SimpleVQA	52.2	58.3
SkillsBench Avg5	12.3	4.4
supergpqa	61.4	63.4
SWE-bench Multilingual	43.4	60.3
SWE-bench Pro	13.8	44.6
SWE-bench Verified	57.4	70
TauBench V3 - Banking	12	4.9
Terminal-Bench 2.0	34.2	40.5
Terminal-Bench 2.1	39	40.8
Terminal-Bench Hard	25	26.5
tool_decathlon	12	28.7
VITA-Bench	36.9	31.9
WideSearch	38.3	59.1
winogrande	79.1	79.2
ZEROBench_sub	26.3	34.1
τ²-Bench Telecom (AA run)	43.6	89.2
τ³-Bench	59	68.9

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.