Which is better, GLM-5.2 Full Open Source or GPT-5.4?

Across 46 shared benchmarks tracked on Vector Wire, GLM-5.2 Full Open Source leads on 14 and GPT-5.4 leads on 31 — GPT-5.4 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GLM-5.2 Full Open Source vs GPT-5.4

Z.aivsOpenAI46 shared benchmarks14–31 head-to-head

Benchmark	GLM-5.2 Full Open Source	GPT-5.4
AA Agentic Index	45.7	58.2
AA Intelligence	52.6	53.1
AA-LCR	76.7	77.7
AA-Omniscience	4.4	5.8
aime_2026	99.2	99.2
apexAgents	33.7	33.3
arc_agi_1	77	93.7
ARC-AGI-2	22.8	74
arena_elo	1471	1477
arena_text_factuality	1460	1476
Artificial Analysis Coding Index	68.8	71.1
coding_arena_elo	1582	1457
critpt	20.9	23.4
DeepSWE 1.1	44	52
gdpval	50.3	50.1
GPQA Diamond	91.2	93
HiL-Bench	43.7	9.7
HLE	54.7	43.7
HLE (with tools)	54.7	52.1
hmmt_feb_2026	92.5	97.7
hmmt_nov_2025	94.4	95.8
IFBench	73.3	73.9
imo_answer_bench	91	91.4
itbenchSre	42.7	34.5
livebench_agentic_coding	51.8	53.8
livebench_coding	79.7	77.5
livebench_data_analysis	73.7	79.3
livebench_instruction_following	62.3	70.2
livebench_language	76.2	82.6
livebench_math	89.8	94.2
livebench_reasoning	78.6	88.1
MCP Atlas	82.6	70.6
nl2repo	48.9	41.3
officeqa_pro	41.4	51.1
OmniScience Accuracy	24.3	50.9
OmniScience Non-Hallucination	73.7	17.4
scicode	50.5	56.6
simpleqa_verified	38.1	45.3
SWE-bench Pro	62.1	59.1
TauBench V3 - Banking	34.6	39.6
Terminal-Bench 2.1	82.7	78.3
Terminal-Bench Hard	50.8	57.6
tool_decathlon	48.2	54.6
toolathlon	48.2	54.6
τ²-Bench Telecom (AA run)	99.1	87.1
τ³-Bench	26.8	72.9

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.