Which is better, Gemma 4 26B A4B or Qwen3.5 27B?

Across 61 shared benchmarks tracked on Vector Wire, Gemma 4 26B A4B leads on 7 and Qwen3.5 27B leads on 54 — Qwen3.5 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 26B A4B vs Qwen3.5 27B

GooglevsAlibaba61 shared benchmarks7–54 head-to-head

Benchmark	Gemma 4 26B A4B	Qwen3.5 27B
AA Agentic Index	32.1	54.6
AA Intelligence	26.1	34.6
AA-LCR	61.7	72.3
AA-Omniscience	-50.8	-44
AI2D	88.3	92.9
aime_2026	88.3	92.6
arena_vision	1238	1221
Artificial Analysis Coding Index	39.3	34.9
browsecomp	26.3	61
C-Eval	82.5	90.5
CC-OCR	74.5	81
charxiv_rq	69	79.5
Claw Eval (pass@3)	28	46.2
Claw-Eval Avg	58.8	64.3
coding_arena_elo	1361	1357
critpt	0	0.9
DeepPlanning	16.2	22.6
gdpval	25.7	33
GPQA Diamond	82.3	85.8
HallusionBench	66.1	70
HLE	19.3	48.5
hmmt_feb_2025	91.7	92
hmmt_feb_2026	79	84.3
hmmt_nov_2025	87.5	89.8
IFBench	72.4	76.5
ifeval	96.4	95
imo_answer_bench	74.3	79.9
itbenchSre	23.6	35.5
LiveCodeBench v6	77.1	80.7
mathvision	82.4	86
MathVista	79.4	87.8
MCP Atlas	50	68.4
mcpmark	14.2	36.3
MMBench EN-DEV-v1.1	89	92.6
mmlu_redux	92.7	93.2
MMLU-Pro	85.2	86.1
mmmlu	86.3	85.9
MMMU	78.4	82.3
MMMU-Pro	73.8	75
nl2repo	11.6	27.3
OmniDocBench 1.5	74.4	88.9
OmniScience Accuracy	19.1	20.7
OmniScience Non-Hallucination	13.6	24.9
QwenClawBench	38.7	52.2
QwenWebBench	1178	1068
RealWorldQA	72.2	83.7
scicode	40.3	39.5
SimpleVQA	52.2	56
SkillsBench Avg5	12.3	27.2
supergpqa	61.4	65.6
SWE-bench Multilingual	43.4	69.3
SWE-bench Pro	13.8	51.2
SWE-bench Verified	57.4	75
Terminal-Bench 2.0	34.2	41.6
Terminal-Bench Hard	25	32.6
tool_decathlon	12	31.5
VITA-Bench	36.9	41.9
WideSearch	38.3	66.4
ZEROBench_sub	26.3	36.2
τ²-Bench Telecom (AA run)	43.6	93.9
τ³-Bench	59	68.4

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.