Which is better, Claude Fable 5 or DeepSeek-V4-Pro?

Across 51 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 48 and DeepSeek-V4-Pro leads on 3 — Claude Fable 5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Fable 5 vs DeepSeek-V4-Pro

AnthropicvsDeepSeek51 shared benchmarks48–3 head-to-head

Benchmark	Claude Fable 5	DeepSeek-V4-Pro
AA Agentic Index	56.6	63.3
AA Intelligence	62.1	53
AA-LCR	76.7	70
AA-Omniscience	43.3	-10.6
Agents' Last Exam	25.7	16.5
arena_elo	1508	1457
arena_text_factuality	1486	1450
Artificial Analysis Coding Index	76.5	59.4
browsecomp	88	83.4
BrowseComp (w/ Ctx)	88	83.4
coding_arena_elo	1626	1582
critpt	28.6	13
DeepSWE	70	12.8
FORTRESS (Adversarial)	96	36
FORTRESS (Benign)	55.1	98.5
gdpval	61.9	49
GDPval-AA (Elo)	1932	1554
GDPval-AA v2	1760	1307
Global-MMLU-Lite	93.3	89.3
GPQA Diamond	92.6	90.5
HLE	64.5	48.2
HLE (with tools)	64.5	48.2
IFBench	63.5	76.5
livebench	78.3	73.6
livebench_agentic_coding	62.2	42.6
livebench_coding	86	70
livebench_data_analysis	80.5	74.5
livebench_instruction_following	75.8	62.4
livebench_language	90.7	78.1
livebench_math	96	90.7
livebench_reasoning	89.7	82.7
MCP Atlas	84.7	74.2
OmniScience Accuracy	65.3	43
OmniScience Non-Hallucination	36.4	12.2
Program Bench	76.8	47.8
scicode	60.2	50
simplebench	81.9	50.9
simpleqa_verified	68.3	57.9
strongreject	98.7	98.6
SWE-bench Multilingual	86.6	76.2
SWE-bench Pro	80	55.4
SWE-bench Verified	95	80.6
SWEBench Pro (Public)	80	55.4
Tau 3 Banking	26.8	26
TauBench V3 - Banking	38.1	30.1
Terminal Bench 2.1 (Best Harness)	84.6	64
Terminal-Bench 2.1	88	72.1
Terminal-Bench Hard	62.9	46.2
toolathlon	61.7	51.8
τ²-Bench Telecom (AA run)	98.5	96.2
τ³-Bench	26.8	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.