Which is better, Gemini 3.1 Pro or GLM-5.1?

Across 59 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 41 and GLM-5.1 leads on 18 — Gemini 3.1 Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 3.1 Pro vs GLM-5.1

GooglevsZ.ai59 shared benchmarks41–18 head-to-head

Benchmark	Gemini 3.1 Pro	GLM-5.1
AA Agentic Index	23	66
AA Intelligence	47.7	41
AA-LCR	79	68
AA-Omniscience	32.9	0.8
AgentWorldBench - Android	61.4	59.1
AgentWorldBench - MCP	59.1	67.6
AgentWorldBench - OS	66.9	59.1
AgentWorldBench - Overall	54.6	51.3
AgentWorldBench - Search	30.2	22.5
AgentWorldBench - SWE	59.1	52.1
AgentWorldBench - Terminal	52.5	47.3
AgentWorldBench - Web	52.8	51.5
aime_2026	98.3	95.8
Apex (Pass@1)	60.9	11.5
Apex Shortlist (Pass@1)	89.1	72.4
arena_elo	1486	1468
arena_text_factuality	1471	1458
Artificial Analysis Coding Index	68.8	55.8
browsecomp	85.9	79.3
browsecomp_with_context_manager	85.9	79.3
coding_arena_elo	1447	1510
critpt	17.7	4.6
cybergym	38.8	68.7
DeepSWE	10	18
Finance Agent v2	43	44.8
frontiermath_tier_4	16.7	12.5
gdpval	23.3	49.5
GDPval-AA (Elo)	1317	1535
GPQA Diamond	94.3	86.8
HLE	51.4	52.3
HLE (with tools)	51.6	52.3
hmmt_feb_2026	94.7	89.4
hmmt_nov_2025	94.8	94
IFBench	77.1	76.3
imo_answer_bench	91	83.8
itbenchSre	30.3	40.3
livebench	79.9	70.2
MCP Atlas	78.2	75.6
MCPAtlas Public (Pass@1)	69.2	71.8
MMLU-Pro	91	86
nl2repo	33.4	42.7
OmniScience Accuracy	55.3	25.2
OmniScience Non-Hallucination	50.1	70.1
PostTrainBench	21.6	20.1
Program Bench	39.5	50.9
scicode	59	43.8
simplebench	79.6	55.1
simpleqa_verified	77.3	38.1
SWE-bench Multilingual	76.9	73.3
SWE-bench Pro	54.2	58.4
SWE-Marathon	4	1
TauBench V3 - Banking	21.4	13.6
Terminal-Bench 2.0	68.5	69
Terminal-Bench 2.1	74	63.5
Terminal-Bench Hard	68.5	43.2
tool_decathlon	48.8	40.7
toolathlon	48.8	40.7
τ²-Bench Telecom (AA run)	99.3	97.7
τ³-Bench	67.1	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.