Which is better, Gemma 4 26B A4B or Kimi K2.5?

Across 49 shared benchmarks tracked on Vector Wire, Gemma 4 26B A4B leads on 2 and Kimi K2.5 leads on 47 — Kimi K2.5 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 Kimi K2.5

GooglevsMoonshot49 shared benchmarks2–47 head-to-head

Benchmark	Gemma 4 26B A4B	Kimi K2.5
AA Agentic Index	32.1	52.8
AA Intelligence	26.1	36
AA-LCR	61.7	73
AA-Omniscience	-50.8	-7.3
AIME 2025	90	96.1
aime_2026	88.3	95.8
arena_elo	1438	1450
arena_text_factuality	1445	1445
arena_vision	1238	1247
Artificial Analysis Coding Index	39.3	46.8
browsecomp	26.3	74.9
charxiv_rq	69	78.7
coding_arena_elo	1361	1436
critpt	0	3.1
gdpval	25.7	38.3
GDPval-AA v2	807	1009
GPQA Diamond	82.3	87.9
HLE	19.3	50.2
HLE (with tools)	17.2	51.8
hmmt_feb_2025	91.7	95.4
hmmt_feb_2026	79	87.1
hmmt_nov_2025	87.5	91.1
IFBench	72.4	70.2
imo_answer_bench	74.3	81.8
LiveCodeBench v6	77.1	85
mathvision	82.4	84.2
MathVista	79.4	90.1
MCP Atlas	50	64
mcpmark	14.2	29.5
MMLU-Pro	85.2	87.1
MMMU-Pro	73.8	78.5
nl2repo	11.6	32
OmniDocBench 1.5	74.4	88.8
OmniScience Accuracy	19.1	35.2
OmniScience Non-Hallucination	13.6	50
scicode	40.3	49
SimpleVQA	52.2	71.2
SWE-bench Multilingual	43.4	73
SWE-bench Pro	13.8	53.8
SWE-bench Verified	57.4	76.8
TauBench V3 - Banking	12	14.2
Terminal-Bench 2.0	34.2	50.8
Terminal-Bench 2.1	39	45.7
Terminal-Bench Hard	25	34.8
tool_decathlon	12	27.8
WideSearch	38.3	79
τ²-Bench	68.2	85.4
τ²-Bench Telecom (AA run)	43.6	95.9
τ³-Bench	59	66

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.