Which is better, Gemma 4 26B A4B or GPT-5.2?

Across 47 shared benchmarks tracked on Vector Wire, Gemma 4 26B A4B leads on 4 and GPT-5.2 leads on 43 — GPT-5.2 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemma 4 26B A4B vs GPT-5.2

GooglevsOpenAI47 shared benchmarks4–43 head-to-head

Benchmark	Gemma 4 26B A4B	GPT-5.2
AA Agentic Index	32.1	60.2
AA Intelligence	26.1	43.3
AA-LCR	61.7	79.3
AA-Omniscience	-50.8	0.3
AIME 2025	90	100
aime_2026	88.3	98.3
arena_elo	1438	1437
arena_text_factuality	1445	1442
arena_vision	1238	1245
Artificial Analysis Coding Index	39.3	48.7
browsecomp	26.3	65.8
charxiv_rq	69	82.1
coding_arena_elo	1361	1418
critpt	0	11.6
gdpval	25.7	48.3
GPQA (unspecified)	82.3	92.4
GPQA Diamond	82.3	92.4
HLE	19.3	45.5
HLE (with tools)	17.2	45.5
hmmt_feb_2025	91.7	100
hmmt_feb_2026	79	97
hmmt_nov_2025	87.5	99.2
IFBench	72.4	75.4
imo_answer_bench	74.3	86.3
LiveCodeBench	79.8	89
mathvision	82.4	83
MathVista	79.4	82.8
MCP Atlas	50	68
MMLU-Pro	85.2	87
mmmlu	86.3	89.6
MMMU-Pro	73.8	79.5
mrcr_v2_8needle_128k_average	44.1	83.8
OmniDocBench 1.5	74.4	85.7
OmniScience Accuracy	19.1	44.3
OmniScience Non-Hallucination	13.6	38.4
scicode	40.3	52.1
SimpleVQA	52.2	55.8
SWE-bench Multilingual	43.4	72
SWE-bench Pro	13.8	55.6
SWE-bench Verified	57.4	80
TauBench V3 - Banking	12	11.1
Terminal-Bench 2.0	34.2	54
Terminal-Bench Hard	25	62.2
tool_decathlon	12	46.3
τ²-Bench	68.2	85.5
τ²-Bench Telecom (AA run)	43.6	98.7
τ³-Bench	59	12

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.