Which is better, Claude Opus 4.6 or GLM-5.1?

Across 52 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 45 and GLM-5.1 leads on 7 — Claude Opus 4.6 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 4.6 vs GLM-5.1

AnthropicvsZ.ai52 shared benchmarks45–7 head-to-head

Benchmark	Claude Opus 4.6	GLM-5.1
AA Agentic Index	67.6	66
AA Intelligence	44.9	41
AA-LCR	74.3	68
AA-Omniscience	13.7	0.8
AgentWorldBench - Android	61.7	59.1
AgentWorldBench - MCP	69.9	67.6
AgentWorldBench - OS	70.2	59.1
AgentWorldBench - Overall	57.8	51.3
AgentWorldBench - Search	29.3	22.5
AgentWorldBench - SWE	64.5	52.1
AgentWorldBench - Terminal	57.5	47.3
AgentWorldBench - Web	51.4	51.5
aime_2026	96.7	95.8
Apex (Pass@1)	34.5	11.5
Apex Shortlist (Pass@1)	85.9	72.4
arena_elo	1497	1468
arena_text_factuality	1487	1458
Artificial Analysis Coding Index	48.1	55.8
browsecomp	86.8	79.3
browsecomp_with_context_manager	84	79.3
coding_arena_elo	1545	1510
critpt	12.6	4.6
cybergym	73.8	68.7
frontiermath_tier_4	22.9	12.5
gdpval	55.9	49.5
GDPval-AA (Elo)	1619	1535
GPQA Diamond	91.3	86.8
HLE	53.1	52.3
HLE (with tools)	62.7	52.3
hmmt_feb_2026	96.2	89.4
hmmt_nov_2025	96.3	94
IFBench	62.5	76.3
imo_answer_bench	75.3	83.8
livebench	76.3	70.2
MCP Atlas	76.8	75.6
MCPAtlas Public (Pass@1)	73.8	71.8
MMLU-Pro	89.1	86
nl2repo	49.8	42.7
OmniScience Accuracy	47	25.2
OmniScience Non-Hallucination	37.2	70.1
scicode	52	43.8
simplebench	67.6	55.1
simpleqa_verified	46.5	38.1
SWE-bench Multilingual	77.8	73.3
SWE-bench Pro	57.3	58.4
Terminal-Bench 2.0	65.4	69
Terminal-Bench Hard	65.4	43.2
tool_decathlon	47.2	40.7
toolathlon	56.8	40.7
Vending-Bench 2	801759	563441
τ²-Bench Telecom (AA run)	99.3	97.7
τ³-Bench	72.4	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.