Which is better, Gemini 3.1 Pro or Gemma 4 26B A4B?

Across 48 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 45 and Gemma 4 26B A4B leads on 3 — Gemini 3.1 Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 3.1 Pro vs Gemma 4 26B A4B

GooglevsGoogle48 shared benchmarks45–3 head-to-head

Benchmark	Gemini 3.1 Pro	Gemma 4 26B A4B
AA Agentic Index	23	32.1
AA Intelligence	47.7	26.1
AA-LCR	79	61.7
AA-Omniscience	32.9	-50.8
aime_2026	98.3	88.3
arena_elo	1486	1438
arena_text_factuality	1471	1445
arena_vision	1280	1238
Artificial Analysis Coding Index	68.8	39.3
browsecomp	85.9	26.3
charxiv_rq	89.9	69
Codeforces ELO	405	1718
coding_arena_elo	1447	1361
critpt	17.7	0
gdpval	23.3	25.7
GDPval-AA v2	962	807
GPQA Diamond	94.3	82.3
HLE	51.4	19.3
HLE (with tools)	51.6	17.2
hmmt_feb_2026	94.7	79
hmmt_nov_2025	94.8	87.5
IFBench	77.1	72.4
imo_answer_bench	91	74.3
itbenchSre	30.3	23.6
LiveCodeBench	91.7	79.8
LiveCodeBench v6	91.7	77.1
mathvision	89.8	82.4
MCP Atlas	78.2	50
mcpmark	55.9	14.2
mmlu	92.6	77.8
MMLU-Pro	91	85.2
mmmlu	92.6	86.3
MMMU-Pro	83	73.8
mrcr_v2_8needle_128k_average	84.9	44.1
nl2repo	33.4	11.6
OmniScience Accuracy	55.3	19.1
OmniScience Non-Hallucination	50.1	13.6
scicode	59	40.3
SWE-bench Multilingual	76.9	43.4
SWE-bench Pro	54.2	13.8
SWE-bench Verified	80.6	57.4
TauBench V3 - Banking	21.4	12
Terminal-Bench 2.0	68.5	34.2
Terminal-Bench 2.1	74	39
Terminal-Bench Hard	68.5	25
tool_decathlon	48.8	12
τ²-Bench Telecom (AA run)	99.3	43.6
τ³-Bench	67.1	59

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.