Which is better, DeepSeek-V3.2 or GLM-5.1?

Across 38 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 3 and GLM-5.1 leads on 34 — GLM-5.1 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V3.2 vs GLM-5.1

DeepSeekvsZ.ai38 shared benchmarks3–34 head-to-head

Benchmark	DeepSeek-V3.2	GLM-5.1
AA Agentic Index	39.8	66
AA Intelligence	32.8	41
AA-LCR	70.7	68
AA-Omniscience	-22.5	0.8
aime_2026	95.1	95.8
arena_text_factuality	1433	1458
Artificial Analysis Coding Index	44.2	55.8
browsecomp	51.4	79.3
browsecomp_with_context_manager	67.6	79.3
coding_arena_elo	1368	1510
critpt	2.9	4.6
cybergym	17.3	68.7
frontiermath_tier_4	2.1	12.5
gdpval	18.8	49.5
GPQA Diamond	84	86.8
HLE	40.8	52.3
HLE (with tools)	40.8	52.3
HMMT 2025	90.2	94
hmmt_feb_2026	84.1	89.4
hmmt_nov_2025	90.2	94
IFBench	61	76.3
imo_answer_bench	78.3	83.8
MCP Atlas	62.2	75.6
MMLU-Pro	86	86
OmniScience Accuracy	33	25.2
OmniScience Non-Hallucination	17.3	70.1
scicode	39	43.8
simpleqa_verified	27.5	38.1
SWE-bench Multilingual	70.2	73.3
SWE-bench Pro	15.6	58.4
TauBench V3 - Banking	18.8	13.6
Terminal-Bench 2.0	46.4	69
Terminal-Bench 2.1	46.8	63.5
Terminal-Bench Hard	35.6	43.2
tool_decathlon	35.2	40.7
toolathlon	35.2	40.7
τ²-Bench Telecom (AA run)	90.6	97.7
τ³-Bench	69.2	70.6

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.