Which is better, Claude Opus 4.6 or DeepSeek-V3.2?

Across 50 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 47 and DeepSeek-V3.2 leads on 3 — 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 DeepSeek-V3.2

AnthropicvsDeepSeek50 shared benchmarks47–3 head-to-head

Benchmark	Claude Opus 4.6	DeepSeek-V3.2
AA Agentic Index	67.6	39.8
AA Intelligence	44.9	32.8
AA-LCR	74.3	70.7
AA-Omniscience	13.7	-22.5
AIME 2025	99.8	93.1
aime_2026	96.7	95.1
AIME25 no tools	95.6	89.3
apexAgents	33	14.5
arc_agi_1	93	57
ARC-AGI-2	68.8	4
arena_text_factuality	1487	1433
Artificial Analysis Coding Index	48.1	44.2
browsecomp	86.8	51.4
browsecomp_with_context_manager	84	67.6
coding_arena_elo	1545	1368
critpt	12.6	2.9
cybergym	73.8	17.3
deepsearchqa_f1	91.3	60.9
frontiermath_tier_4	22.9	2.1
gdpval	55.9	18.8
GPQA Diamond	91.3	84
HLE	53.1	40.8
HLE (with tools)	62.7	40.8
hmmt_feb_2026	96.2	84.1
hmmt_nov_2025	96.3	90.2
IFBench	62.5	61
imo_answer_bench	75.3	78.3
LiveCodeBench	88.8	86
LiveCodeBench v6	88.8	83.3
MCP Atlas	76.8	62.2
mcpmark	56.7	38
MMLU-Pro	89.1	86
OmniScience Accuracy	47	33
OmniScience Non-Hallucination	37.2	17.3
scicode	52	39
simpleqa_verified	46.5	27.5
swe_bench_bash	75.6	60
SWE-bench Multilingual	77.8	70.2
SWE-bench Pro	57.3	15.6
SWE-bench Verified	80.8	73.1
Terminal-Bench 2.0	65.4	46.4
Terminal-Bench Hard	65.4	35.6
tool_decathlon	47.2	35.2
toolathlon	56.8	35.2
vectara_answer_rate	99.8	92.6
vectara_avg_summary_length	137.6	62
vectara_factual_consistency	87.8	93.7
vectara_hallucination_rate ↓	12.2	6.3
τ²-Bench Telecom (AA run)	99.3	90.6
τ³-Bench	72.4	69.2

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.