Which is better, Gemini 3 Pro or Gemma 4 31B?

Across 48 shared benchmarks tracked on Vector Wire, Gemini 3 Pro leads on 43 and Gemma 4 31B leads on 5 — Gemini 3 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 Pro vs Gemma 4 31B

GooglevsGoogle48 shared benchmarks43–5 head-to-head

Benchmark	Gemini 3 Pro	Gemma 4 31B
AA Agentic Index	52	14.4
AA Intelligence	40.6	29.7
AA-LCR	73	68.3
AA-Omniscience	15.3	-47.9
aime_2026	91.7	89.2
arena_elo	1485	1451
arena_text_factuality	1481	1449
arena_vision	1290	1252
Artificial Analysis Coding Index	46.5	43.4
charxiv_rq	81.4	67.9
coding_arena_elo	1438	1364
critpt	9.1	1.4
gdpval	34.2	15.5
GPQA Diamond	91.9	85.7
HLE	45.8	26.5
HLE (with tools)	45.8	26.5
hmmt_feb_2025	97.5	88.7
hmmt_feb_2026	86.4	77.2
hmmt_nov_2025	93.3	87.5
IFBench	70.4	75.6
imo_answer_bench	83.3	74.5
LiveCodeBench v6	90.7	80
mathvision	86.1	85.6
MathVista	89.8	79.3
MCP Atlas	70.3	57.2
MMLU-Pro	90.1	85.2
mmmlu	91.8	88.4
MMMU-Pro	81	76.9
mrcr_v2_8needle_128k_average	77	66.4
OCRBench	90.3	86.1
OmniDocBench 1.5	88.5	80.1
OmniScience Accuracy	55.8	20
OmniScience Non-Hallucination	10	18.1
scicode	56.1	43.4
simpleqa_verified	72.9	9.6
SimpleVQA	69.7	52.9
SWE-bench Multilingual	68.7	51.7
SWE-bench Pro	43.3	35.7
SWE-bench Verified	78	52
Terminal-Bench 2.0	54.2	42.9
Terminal-Bench Hard	56.9	36.4
tool_decathlon	36.4	21.2
vectara_answer_rate	99.4	100
vectara_avg_summary_length	101.9	75.8
vectara_factual_consistency	86.4	92.6
vectara_hallucination_rate ↓	13.6	7.4
τ²-Bench	98	76.9
τ²-Bench Telecom (AA run)	98	65.5

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.