Which is better, Claude Opus 4.8 or GLM-5.1?

Across 53 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 46 and GLM-5.1 leads on 7 — 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-5.1

AnthropicvsZ.ai53 shared benchmarks46–7 head-to-head

Benchmark	Claude Opus 4.8	GLM-5.1
AA Agentic Index	49.4	66
AA Intelligence	57.3	41
AA-LCR	73	68
AA-Omniscience	28.8	0.8
AgentWorldBench - Android	61.5	59.1
AgentWorldBench - MCP	54.9	67.6
AgentWorldBench - OS	66.6	59.1
AgentWorldBench - Overall	56.6	51.3
AgentWorldBench - Search	35.1	22.5
AgentWorldBench - SWE	64.1	52.1
AgentWorldBench - Terminal	59.2	47.3
AgentWorldBench - Web	54.7	51.5
aime_2026	100	95.8
arena_elo	1482	1468
arena_text_factuality	1464	1458
Artificial Analysis Coding Index	74.3	55.8
browsecomp	88.5	79.3
coding_arena_elo	1539	1510
critpt	20.9	4.6
cybergym	83.1	68.7
DeepSWE	59	18
Finance Agent v2	53.9	44.8
frontiermath_tier_4	31.3	12.5
gdpval	54.2	49.5
GDPval-AA (Elo)	1890	1535
GPQA Diamond	93.6	86.8
HLE	57.9	52.3
HLE (with tools)	57.9	52.3
hmmt_feb_2026	96.7	89.4
hmmt_nov_2025	96.5	94
IFBench	62.2	76.3
imo_answer_bench	83.5	83.8
livebench	77.2	70.2
MCP Atlas	83.6	75.6
nl2repo	69.7	42.7
OmniScience Accuracy	48.8	25.2
OmniScience Non-Hallucination	60.7	70.1
PostTrainBench	37.2	20.1
Program Bench	71.9	50.9
scicode	53.5	43.8
simplebench	64.8	55.1
simpleqa_verified	39.5	38.1
SWE-bench Multilingual	84.4	73.3
SWE-bench Pro	69.2	58.4
SWE-Marathon	40	1
TauBench V3 - Banking	34.2	13.6
Terminal-Bench 2.0	74.6	69
Terminal-Bench 2.1	85	63.5
Terminal-Bench Hard	58.3	43.2
tool_decathlon	59.9	40.7
toolathlon	59.9	40.7
τ²-Bench Telecom (AA run)	94.4	97.7
τ³-Bench	27.6	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.