Which is better, Claude Opus 4.8 or DeepSeek-V4-Pro?

Across 66 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 58 and DeepSeek-V4-Pro leads on 8 — 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 DeepSeek-V4-Pro

AnthropicvsDeepSeek66 shared benchmarks58–8 head-to-head

Benchmark	Claude Opus 4.8	DeepSeek-V4-Pro
AA Agentic Index	49.4	63.3
AA Intelligence	57.3	53
AA-LCR	73	70
AA-Omniscience	28.8	-10.6
Agents' Last Exam	27	16.5
AgentWorldBench - Android	61.5	55.2
AgentWorldBench - MCP	54.9	63.3
AgentWorldBench - OS	66.6	63.7
AgentWorldBench - Overall	56.6	53
AgentWorldBench - Search	35.1	27.6
AgentWorldBench - SWE	64.1	59.4
AgentWorldBench - Terminal	59.2	51.3
AgentWorldBench - Web	54.7	50.3
aime_2026	100	96.7
arena_elo	1482	1457
arena_text_factuality	1464	1450
Artificial Analysis Coding Index	74.3	59.4
AutomationBench (Public)	27.2	12.8
AutomationBench Public	27.2	12.8
browsecomp	88.5	83.4
coding_arena_elo	1539	1582
critpt	20.9	13
cybergym	83.1	52.7
DeepSWE	59	12.8
DSBench-FullStack	71.6	41.8
DSBench-Hard	71.7	31.1
gdpval	54.2	49
GDPval-AA (Elo)	1890	1554
GDPval-AA v2	1593	1307
GPQA Diamond	93.6	90.5
HLE	57.9	48.2
HLE (with tools)	57.9	48.2
HLE (wo / w tools)	57.9	48.2
hmmt_feb_2026	96.7	95.2
hmmt_nov_2025	96.5	94.4
IFBench	62.2	76.5
imo_answer_bench	83.5	89.8
livebench	77.2	73.6
livebench_agentic_coding	50.5	42.6
livebench_coding	81.8	70
livebench_data_analysis	66	74.5
livebench_instruction_following	72	62.4
livebench_language	79.7	78.1
livebench_math	94.3	90.7
livebench_reasoning	89.2	82.7
MCP Atlas	83.6	74.2
nl2repo	69.7	38.5
OmniScience Accuracy	48.8	43
OmniScience Non-Hallucination	60.7	12.2
Program Bench	71.9	47.8
scicode	53.5	50
simplebench	64.8	50.9
simpleqa_verified	39.5	57.9
SWE-bench Multilingual	84.4	76.2
SWE-bench Pro	69.2	55.4
SWE-bench Verified	88.6	80.6
Tau 3 Banking	27.6	26
TauBench V3 - Banking	34.2	30.1
Terminal-Bench 2.0	74.6	67.9
Terminal-Bench 2.1	85	72.1
Terminal-Bench Hard	58.3	46.2
tool_decathlon	59.9	52.8
toolathlon	59.9	51.8
usamo_2026	96.7	60.7
τ²-Bench Telecom (AA run)	94.4	96.2
τ³-Bench	27.6	25.8

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.