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

Across 55 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 52 and Gemma 4 26B A4B leads on 3 — Claude Opus 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 Opus 4.5 vs Gemma 4 26B A4B

AnthropicvsGoogle55 shared benchmarks52–3 head-to-head

Benchmark	Claude Opus 4.5	Gemma 4 26B A4B
AA Agentic Index	59.6	32.1
AA Intelligence	41.9	26.1
AA-LCR	76	61.7
AA-Omniscience	14	-50.8
AIME 2025	92.8	90
aime_2026	95.1	88.3
arena_elo	1473	1438
arena_text_factuality	1458	1445
Artificial Analysis Coding Index	47.8	39.3
browsecomp	37	26.3
C-Eval	92.2	82.5
CC-OCR	76.9	74.5
charxiv_rq	68.5	69
Claw Eval (pass@3)	59.6	28
Claw-Eval Avg	76.6	58.8
coding_arena_elo	1495	1361
critpt	4.6	0
gdpval	47.3	25.7
GPQA Diamond	87	82.3
HLE	30.8	19.3
HLE (with tools)	43.4	17.2
hmmt_feb_2025	92.9	91.7
hmmt_feb_2026	85.3	79
hmmt_nov_2025	93.3	87.5
IFBench	58	72.4
imo_answer_bench	84	74.3
LiveCodeBench	87	79.8
LiveCodeBench v6	84.8	77.1
mathvision	77.1	82.4
MathVista	80.2	79.4
MCP Atlas	69.8	50
mmlu_redux	95.6	92.7
MMLU-Pro	90	85.2
mmmlu	90.8	86.3
MMMU	80.7	78.4
MMMU-Pro	74	73.8
nl2repo	43.2	11.6
OmniDocBench 1.5	87.7	74.4
OmniScience Accuracy	46.6	19.1
OmniScience Non-Hallucination	39	13.6
QwenClawBench	52.3	38.7
QwenWebBench	1536	1178
RealWorldQA	77	72.2
scicode	50	40.3
SimpleVQA	69.7	52.2
SkillsBench Avg5	45.3	12.3
supergpqa	70.6	61.4
SWE-bench Multilingual	77.5	43.4
SWE-bench Pro	57.1	13.8
SWE-bench Verified	81.5	57.4
Terminal-Bench 2.0	59.3	34.2
Terminal-Bench Hard	47	25
tool_decathlon	43.5	12
τ²-Bench	98.2	68.2
τ²-Bench Telecom (AA run)	89.5	43.6

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.