Which is better, DeepSeek-V4-Pro or GLM-5.1?

Across 54 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 32 and GLM-5.1 leads on 22 — DeepSeek-V4-Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V4-Pro vs GLM-5.1

DeepSeekvsZ.ai54 shared benchmarks32–22 head-to-head

Benchmark	DeepSeek-V4-Pro	GLM-5.1
AA Agentic Index	63.3	66
AA Intelligence	53	41
AA-LCR	70	68
AA-Omniscience	-10.6	0.8
AgentWorldBench - Android	55.2	59.1
AgentWorldBench - MCP	63.3	67.6
AgentWorldBench - OS	63.7	59.1
AgentWorldBench - Overall	53	51.3
AgentWorldBench - Search	27.6	22.5
AgentWorldBench - SWE	59.4	52.1
AgentWorldBench - Terminal	51.3	47.3
AgentWorldBench - Web	50.3	51.5
aime_2026	96.7	95.8
Apex (Pass@1)	38.3	11.5
Apex Shortlist (Pass@1)	90.2	72.4
arena_elo	1457	1468
arena_text_factuality	1450	1458
Artificial Analysis Coding Index	59.4	55.8
browsecomp	83.4	79.3
coding_arena_elo	1582	1510
critpt	13	4.6
cybergym	52.7	68.7
DeepSWE	12.8	18
gdpval	49	49.5
GDPval-AA (Elo)	1554	1535
GPQA Diamond	90.5	86.8
HLE	48.2	52.3
HLE (with tools)	48.2	52.3
hmmt_feb_2026	95.2	89.4
hmmt_nov_2025	94.4	94
IFBench	76.5	76.3
imo_answer_bench	89.8	83.8
itbenchSre	38.3	40.3
livebench	73.6	70.2
MCP Atlas	74.2	75.6
MCPAtlas Public (Pass@1)	73.6	71.8
MMLU-Pro	87.5	86
nl2repo	38.5	42.7
OmniScience Accuracy	43	25.2
OmniScience Non-Hallucination	12.2	70.1
Program Bench	47.8	50.9
scicode	50	43.8
simplebench	50.9	55.1
simpleqa_verified	57.9	38.1
SWE-bench Multilingual	76.2	73.3
SWE-bench Pro	55.4	58.4
TauBench V3 - Banking	30.1	13.6
Terminal-Bench 2.0	67.9	69
Terminal-Bench 2.1	72.1	63.5
Terminal-Bench Hard	46.2	43.2
tool_decathlon	52.8	40.7
toolathlon	51.8	40.7
τ²-Bench Telecom (AA run)	96.2	97.7
τ³-Bench	25.8	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.