Which is better, Gemma 4 31B or Kimi K2.6?

Across 45 shared benchmarks tracked on Vector Wire, Gemma 4 31B leads on 5 and Kimi K2.6 leads on 40 — Kimi K2.6 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 31B vs Kimi K2.6

GooglevsMoonshot45 shared benchmarks5–40 head-to-head

Benchmark	Gemma 4 31B	Kimi K2.6
AA Agentic Index	14.4	58.7
AA Intelligence	29.7	45.1
AA-LCR	68.3	76.7
AA-Omniscience	-47.9	6.4
aime_2026	89.2	96.4
arena_elo	1451	1461
arena_text_factuality	1449	1455
arena_vision	1252	1264
Artificial Analysis Coding Index	43.4	61.8
charxiv_rq	67.9	86.7
Claw Eval (pass@3)	25	80.9
coding_arena_elo	1364	1509
critpt	1.4	8
gdpval	15.5	41.4
GPQA Diamond	85.7	91.1
HLE	26.5	37.5
HLE (with tools)	26.5	54
hmmt_feb_2026	77.2	94.7
IFBench	75.6	76
imo_answer_bench	74.5	86
itbenchSre	37.3	31.2
LiveCodeBench v6	80	89.6
mathvision	85.6	93.2
MCP Atlas	57.2	68.1
mcpmark	18.1	55.9
MMLU-Pro	85.2	87.1
MMMU-Pro	76.9	80.1
OmniScience Accuracy	20	32.8
OmniScience Non-Hallucination	18.1	60.7
scicode	43.4	53.5
simpleqa_verified	9.6	38.7
SWE-bench Multilingual	51.7	76.7
SWE-bench Pro	35.7	58.6
SWE-bench Verified	52	80.2
TauBench V3 - Banking	14.8	23.3
Terminal-Bench 2.0	42.9	66.7
Terminal-Bench 2.1	43.4	65.9
Terminal-Bench Hard	36.4	43.9
vectara_answer_rate	100	99.7
vectara_avg_summary_length	75.8	116.7
vectara_factual_consistency	92.6	89.2
vectara_hallucination_rate ↓	7.4	10.8
WideSearch	35.2	80.8
τ²-Bench Telecom (AA run)	65.5	95.9
τ³-Bench	67.5	20.6

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.