Which is better, GLM-5.1 or Kimi K2.6?

Across 50 shared benchmarks tracked on Vector Wire, GLM-5.1 leads on 19 and Kimi K2.6 leads on 31 — 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 —

GLM-5.1 vs Kimi K2.6

Z.aivsMoonshot50 shared benchmarks19–31 head-to-head

Benchmark	GLM-5.1	Kimi K2.6
AA Agentic Index	66	58.7
AA Intelligence	41	45.1
AA-LCR	68	76.7
AA-Omniscience	0.8	6.4
AgentWorldBench - Android	59.1	58.9
AgentWorldBench - MCP	67.6	65.2
AgentWorldBench - OS	59.1	60.8
AgentWorldBench - Overall	51.3	53.4
AgentWorldBench - Search	22.5	27.5
AgentWorldBench - SWE	52.1	58.8
AgentWorldBench - Terminal	47.3	52.5
AgentWorldBench - Web	51.5	50.2
aime_2026	95.8	96.4
Apex (Pass@1)	11.5	24
Apex Shortlist (Pass@1)	72.4	75.5
arena_elo	1468	1461
arena_text_factuality	1458	1455
Artificial Analysis Coding Index	55.8	61.8
browsecomp	79.3	86.3
coding_arena_elo	1510	1509
critpt	4.6	8
Finance Agent v2	44.8	44.9
frontiermath_tier_4	12.5	14.6
gdpval	49.5	41.4
GDPval-AA (Elo)	1535	1482
GPQA Diamond	86.8	91.1
HLE	52.3	37.5
HLE (with tools)	52.3	54
hmmt_feb_2026	89.4	94.7
IFBench	76.3	76
imo_answer_bench	83.8	86
itbenchSre	40.3	31.2
livebench	70.2	72.2
MCP Atlas	75.6	68.1
MCPAtlas Public (Pass@1)	71.8	66.6
MMLU-Pro	86	87.1
OmniScience Accuracy	25.2	32.8
OmniScience Non-Hallucination	70.1	60.7
scicode	43.8	53.5
simpleqa_verified	38.1	38.7
SWE-bench Multilingual	73.3	76.7
SWE-bench Pro	58.4	58.6
SWE-QA	72.7	71.6
TauBench V3 - Banking	13.6	23.3
Terminal-Bench 2.0	69	66.7
Terminal-Bench 2.1	63.5	65.9
Terminal-Bench Hard	43.2	43.9
toolathlon	40.7	50
τ²-Bench Telecom (AA run)	97.7	95.9
τ³-Bench	70.6	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.