Which is better, Gemini 3.1 Pro or GLM-5.2 Full Open Source?

Across 60 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 28 and GLM-5.2 Full Open Source leads on 31 — GLM-5.2 Full Open Source 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-5.2 Full Open Source

GooglevsZ.ai60 shared benchmarks28–31 head-to-head

Benchmark	Gemini 3.1 Pro	GLM-5.2 Full Open Source
AA Agentic Index	23	45.7
AA Intelligence	47.7	52.6
AA-LCR	79	76.7
AA-Omniscience	32.9	4.4
aime_2026	98.3	99.2
apex_agents	33.5	35.6
apexAgents	33.5	33.7
arc_agi_1	98	77
ARC-AGI-2	77.1	22.8
arena_elo	1486	1471
arena_text_factuality	1471	1460
Artificial Analysis Coding Index	68.8	68.8
coding_arena_elo	1447	1582
critpt	17.7	20.9
DeepSWE	10	46.2
DeepSWE 1.1	12	44
FORTRESS (Adversarial)	65.2	71.3
FORTRESS (Benign)	98	90
gdpval	23.3	50.3
GDPval-AA v2	962	1514
Global-MMLU-Lite	92.7	89.2
GPQA Diamond	94.3	91.2
HiL-Bench	35.3	43.7
HLE	51.4	54.7
HLE (with tools)	51.6	54.7
hmmt_feb_2026	94.7	92.5
hmmt_nov_2025	94.8	94.4
IFBench	77.1	73.3
imo_answer_bench	91	91
itbenchSre	30.3	42.7
livebench_agentic_coding	44.1	51.8
livebench_coding	76.5	79.7
livebench_data_analysis	78.5	73.7
livebench_instruction_following	79.1	62.3
livebench_language	85.4	76.2
livebench_math	91	89.8
livebench_reasoning	84	78.6
MCP Atlas	78.2	82.6
nl2repo	33.4	48.9
officeqa_pro	42.9	41.4
OmniScience Accuracy	55.3	24.3
OmniScience Non-Hallucination	50.1	73.7
PostTrainBench	21.6	34.3
Program Bench	39.5	63.7
scicode	59	50.5
simplebench	79.6	58.8
simpleqa_verified	77.3	38.1
strongreject	98	98.5
SWE-bench Pro	54.2	62.1
SWE-Marathon	4	13
SWEBench Pro (Public)	54.2	62.1
Tau 3 Banking	16.5	26.8
TauBench V3 - Banking	21.4	34.6
Terminal Bench 2.1 (Best Harness)	73.8	82.7
Terminal-Bench 2.1	74	82.7
Terminal-Bench Hard	68.5	50.8
tool_decathlon	48.8	48.2
toolathlon	48.8	48.2
τ²-Bench Telecom (AA run)	99.3	99.1
τ³-Bench	67.1	26.8

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.