Which is better, Claude Sonnet 4.5 or Gemma 4 31B?

Across 53 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 29 and Gemma 4 31B leads on 23 — Claude Sonnet 4.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Sonnet 4.5 vs Gemma 4 31B

AnthropicvsGoogle53 shared benchmarks29–23 head-to-head

Benchmark	Claude Sonnet 4.5	Gemma 4 31B
AA Agentic Index	50.6	14.4
AA Intelligence	37.4	29.7
AA-LCR	68.3	68.3
AA-Omniscience	-0.1	-47.9
AI2D	87	89
arena_elo	1456	1451
arena_text_factuality	1448	1449
Artificial Analysis Coding Index	52.1	43.4
CC-OCR	68.1	75.7
charxiv_rq	67.2	67.9
coding_arena_elo	1392	1364
critpt	1.1	1.4
gdpval	40.4	15.5
GPQA (unspecified)	83.4	84.3
GPQA Diamond	83.4	85.7
HallusionBench	59.9	67.4
HLE	19.8	26.5
HLE (with tools)	32	26.5
hmmt_feb_2025	79.2	88.7
hmmt_nov_2025	81.7	87.5
IFBench	57.3	75.6
imo_answer_bench	65.9	74.5
LiveCodeBench v6	64	80
MathVista	79.8	79.3
MCP Atlas	59.5	57.2
MMBench EN-DEV-v1.1	88.3	90.9
mmlu_redux	95.6	93.7
MMLU-Pro	88.2	85.2
mmmlu	89.1	88.4
MMMU	79.6	80.4
MMMU-Pro	68.7	76.9
OmniDocBench 1.5	85.8	80.1
OmniScience Accuracy	32.9	20
OmniScience Non-Hallucination	50.9	18.1
RealWorldQA	70.3	72.3
scicode	45	43.4
simpleqa_verified	23.6	9.6
SimpleVQA	57.6	52.9
SWE-bench Multilingual	68	51.7
SWE-bench Pro	43.6	35.7
SWE-bench Verified	82	52
TauBench V3 - Banking	24.5	14.8
Terminal-Bench 2.0	50	42.9
Terminal-Bench 2.1	55.8	43.4
Terminal-Bench Hard	35.6	36.4
vectara_answer_rate	95.6	100
vectara_avg_summary_length	127.8	75.8
vectara_factual_consistency	88	92.6
vectara_hallucination_rate ↓	12	7.4
ZEROBench_sub	26.3	26
τ²-Bench	98	76.9
τ²-Bench Telecom (AA run)	78.1	65.5
τ³-Bench	19	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.