Which is better, Claude Opus 4.6 or DeepSeek-V4-Flash?

Across 51 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 38 and DeepSeek-V4-Flash leads on 12 — 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-V4-Flash

AnthropicvsDeepSeek51 shared benchmarks38–12 head-to-head

Benchmark	Claude Opus 4.6	DeepSeek-V4-Flash
AA Agentic Index	67.6	61.3
AA Intelligence	44.9	51.8
AA-LCR	74.3	74.3
AA-Omniscience	13.7	-14.3
aime_2026	96.7	95.8
Apex (Pass@1)	34.5	33
Apex Shortlist (Pass@1)	85.9	85.7
arc_agi_1	93	84
ARC-AGI-2	68.8	46
arena_elo	1497	1439
arena_text_factuality	1487	1434
Artificial Analysis Coding Index	48.1	69.1
browsecomp	86.8	73.2
Chinese SimpleQA (C-SimpleQA)	76.4	78.9
coding_arena_elo	1545	1431
CorpusQA 1M (ACC)	71.7	60.5
critpt	12.6	16.6
cybergym	73.8	76.7
gdpval	55.9	52.9
GDPval-AA (Elo)	1619	1395
GPQA Diamond	91.3	90.8
HLE	53.1	38.6
HLE (with tools)	62.7	45.1
hmmt_feb_2026	96.2	94.8
IFBench	62.5	79.2
imo_answer_bench	75.3	88.4
livebench_agentic_coding	49	46.8
livebench_coding	78.2	75
livebench_data_analysis	69.9	79.3
livebench_instruction_following	63.3	65.5
livebench_language	83.3	79.2
livebench_math	89.3	86.8
livebench_reasoning	88.7	86.6
LiveCodeBench	88.8	91.6
MCP Atlas	76.8	69
MMLU-Pro	89.1	86.4
MRCR 1M (MMR)	92.9	78.7
nl2repo	49.8	54.2
OmniScience Accuracy	47	40.4
OmniScience Non-Hallucination	37.2	15.6
scicode	52	49.9
simplebench	67.6	46.3
simpleqa_verified	46.5	45
SWE-bench Multilingual	77.8	73.3
SWE-bench Pro	57.3	52.6
SWE-bench Verified	80.8	79
Terminal-Bench 2.0	65.4	56.9
Terminal-Bench Hard	65.4	38.6
toolathlon	56.8	70.3
τ²-Bench Telecom (AA run)	99.3	95.6
τ³-Bench	72.4	22.9

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.