Which is better, Claude Opus 4.8 or GLM-5.2 Full Open Source?

Across 63 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 52 and GLM-5.2 Full Open Source leads on 10 — Claude Opus 4.8 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.8 vs GLM-5.2 Full Open Source

AnthropicvsZ.ai63 shared benchmarks52–10 head-to-head

Benchmark	Claude Opus 4.8	GLM-5.2 Full Open Source
AA Agentic Index	49.4	45.7
AA Intelligence	57.3	52.6
AA-LCR	73	76.7
AA-Omniscience	28.8	4.4
Agents' Last Exam	27	23.8
aime_2026	100	99.2
apex_agents	39.4	35.6
arc_agi_1	92.5	77
ARC-AGI-2	72.1	22.8
arena_elo	1482	1471
arena_text_factuality	1464	1460
Artificial Analysis Coding Index	74.3	68.8
AutomationBench (Public)	27.2	12.9
AutomationBench Public	27.2	12.9
coding_arena_elo	1539	1582
critpt	20.9	20.9
DeepSWE	59	46.2
DeepSWE 1.1	59	44
DSBench-FullStack	71.6	61.8
DSBench-Hard	71.7	54.5
gdpval	54.2	50.3
GDPval-AA v2	1593	1514
GDPval-AA v2 (Elo)	1593	1510
GPQA Diamond	93.6	91.2
HiL-Bench	35.3	43.7
HLE	57.9	54.7
HLE (with tools)	57.9	54.7
HLE (wo / w tools)	57.9	54.7
hmmt_feb_2026	96.7	92.5
hmmt_nov_2025	96.5	94.4
IFBench	62.2	73.3
imo_answer_bench	83.5	91
JobBench	48.4	43.4
livebench_agentic_coding	50.5	51.8
livebench_coding	81.8	79.7
livebench_data_analysis	66	73.7
livebench_instruction_following	72	62.3
livebench_language	79.7	76.2
livebench_math	94.3	89.8
livebench_reasoning	89.2	78.6
MCP Atlas	83.6	82.6
MLS Bench Lite	42.8	40.4
nl2repo	69.7	48.9
officeqa_pro	66.2	41.4
OmniScience Accuracy	48.8	24.3
OmniScience Non-Hallucination	60.7	73.7
PostTrainBench	37.2	34.3
Program Bench	71.9	63.7
ResearchRubrics	73.5	71.1
scicode	53.5	50.5
simplebench	64.8	58.8
simpleqa_verified	39.5	38.1
SpreadsheetBench 2	31.6	28.1
SWE-bench Pro	69.2	62.1
SWE-Marathon	40	13
Tau 3 Banking	27.6	26.8
TauBench V3 - Banking	34.2	34.6
Terminal-Bench 2.1	85	82.7
Terminal-Bench Hard	58.3	50.8
tool_decathlon	59.9	48.2
toolathlon	59.9	48.2
τ²-Bench Telecom (AA run)	94.4	99.1
τ³-Bench	27.6	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.