Which is better, Claude Sonnet 4.5 or Gemma 4 26B A4B?

Across 51 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 34 and Gemma 4 26B A4B leads on 16 — 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 26B A4B

AnthropicvsGoogle51 shared benchmarks34–16 head-to-head

Benchmark	Claude Sonnet 4.5	Gemma 4 26B A4B
AA Agentic Index	50.6	32.1
AA Intelligence	37.4	26.1
AA-LCR	68.3	61.7
AA-Omniscience	-0.1	-50.8
AI2D	87	88.3
AIME 2025	88	90
arena_elo	1456	1438
arena_text_factuality	1448	1445
Artificial Analysis Coding Index	52.1	39.3
browsecomp	24.1	26.3
CC-OCR	68.1	74.5
charxiv_rq	67.2	69
coding_arena_elo	1392	1361
critpt	1.1	0
gdpval	40.4	25.7
GPQA (unspecified)	83.4	82.3
GPQA Diamond	83.4	82.3
HallusionBench	59.9	66.1
HLE	19.8	19.3
HLE (with tools)	32	17.2
hmmt_feb_2025	79.2	91.7
hmmt_nov_2025	81.7	87.5
IFBench	57.3	72.4
imo_answer_bench	65.9	74.3
LiveCodeBench	71	79.8
LiveCodeBench v6	64	77.1
MathVista	79.8	79.4
MCP Atlas	59.5	50
MMBench EN-DEV-v1.1	88.3	89
mmlu_redux	95.6	92.7
MMLU-Pro	88.2	85.2
mmmlu	89.1	86.3
MMMU	79.6	78.4
MMMU-Pro	68.7	73.8
OmniDocBench 1.5	85.8	74.4
OmniScience Accuracy	32.9	19.1
OmniScience Non-Hallucination	50.9	13.6
RealWorldQA	70.3	72.2
scicode	45	40.3
SimpleVQA	57.6	52.2
SWE-bench Multilingual	68	43.4
SWE-bench Pro	43.6	13.8
SWE-bench Verified	82	57.4
TauBench V3 - Banking	24.5	12
Terminal-Bench 2.0	50	34.2
Terminal-Bench 2.1	55.8	39
Terminal-Bench Hard	35.6	25
ZEROBench_sub	26.3	26.3
τ²-Bench	98	68.2
τ²-Bench Telecom (AA run)	78.1	43.6
τ³-Bench	19	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.