Which is better, DeepSeek-V4-Pro or Opus 5?

Across 34 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 2 and Opus 5 leads on 32 — Opus 5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V4-Pro vs Opus 5

DeepSeekvsAnthropic34 shared benchmarks2–32 head-to-head

Benchmark	DeepSeek-V4-Pro	Opus 5
AA Agentic Index	63.3	59.2
AA Intelligence	53	63.1
AA-LCR	70	78.7
AA-Omniscience	-10.6	37.1
arena_elo	1457	1487
arena_text_factuality	1450	1481
Artificial Analysis Coding Index	59.4	78
browsecomp	83.4	90.8
coding_arena_elo	1582	1662
critpt	13	29.1
gdpval	49	67.2
GDPval-AA (Elo)	1554	1827
GDPval-AA v2	1307	1861
GPQA Diamond	90.5	93.7
HLE	48.2	64.7
HLE (with tools)	48.2	64.7
livebench_agentic_coding	42.6	65.2
livebench_coding	70	81.5
livebench_data_analysis	74.5	74.5
livebench_instruction_following	62.4	63.8
livebench_language	78.1	88.7
livebench_math	90.7	95.7
livebench_reasoning	82.7	91.2
MCP Atlas	74.2	85.8
OmniScience Accuracy	43	60.9
OmniScience Non-Hallucination	12.2	40.5
scicode	50	55.7
simplebench	50.9	80.6
simpleqa_verified	57.9	56.7
SWE-bench Multilingual	76.2	89.5
SWE-bench Pro	55.4	79.2
SWE-bench Verified	80.6	96
TauBench V3 - Banking	30.1	44.7
Terminal-Bench 2.1	72.1	89.1

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.