Which is better, Gemini 3.1 Pro or Gemma 4 31B?

Across 49 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 43 and Gemma 4 31B leads on 6 — 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 31B

GooglevsGoogle49 shared benchmarks43–6 head-to-head

Benchmark	Gemini 3.1 Pro	Gemma 4 31B
AA Agentic Index	23	14.4
AA Intelligence	47.7	29.7
AA-LCR	79	68.3
AA-Omniscience	32.9	-47.9
aime_2026	98.3	89.2
arena_elo	1486	1451
arena_text_factuality	1471	1449
arena_vision	1280	1252
Artificial Analysis Coding Index	68.8	43.4
charxiv_rq	89.9	67.9
Codeforces ELO	405	2150
coding_arena_elo	1447	1364
critpt	17.7	1.4
gdpval	23.3	15.5
GPQA Diamond	94.3	85.7
HLE	51.4	26.5
HLE (with tools)	51.6	26.5
hmmt_feb_2026	94.7	77.2
hmmt_nov_2025	94.8	87.5
IFBench	77.1	75.6
imo_answer_bench	91	74.5
itbenchSre	30.3	37.3
LiveCodeBench v6	91.7	80
mathvision	89.8	85.6
MCP Atlas	78.2	57.2
mcpmark	55.9	18.1
MMLU-Pro	91	85.2
mmmlu	92.6	88.4
MMMU-Pro	83	76.9
mrcr_v2_8needle_128k_average	84.9	66.4
nl2repo	33.4	15.5
OmniScience Accuracy	55.3	20
OmniScience Non-Hallucination	50.1	18.1
scicode	59	43.4
simpleqa_verified	77.3	9.6
SWE-bench Multilingual	76.9	51.7
SWE-bench Pro	54.2	35.7
SWE-bench Verified	80.6	52
TauBench V3 - Banking	21.4	14.8
Terminal-Bench 2.0	68.5	42.9
Terminal-Bench 2.1	74	43.4
Terminal-Bench Hard	68.5	36.4
tool_decathlon	48.8	21.2
vectara_answer_rate	99.4	100
vectara_avg_summary_length	107.7	75.8
vectara_factual_consistency	89.6	92.6
vectara_hallucination_rate ↓	10.4	7.4
τ²-Bench Telecom (AA run)	99.3	65.5
τ³-Bench	67.1	67.5

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.