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

Across 46 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 42 and DeepSeek-V4-Flash leads on 4 — 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-Flash

AnthropicvsDeepSeek46 shared benchmarks42–4 head-to-head

Benchmark	Claude Fable 5	DeepSeek-V4-Flash
AA Agentic Index	56.6	61.3
AA Intelligence	62.1	51.8
AA-LCR	76.7	74.3
AA-Omniscience	43.3	-14.3
Agents' Last Exam	25.7	25.2
arc_agi_1	90.5	84
ARC-AGI-2	76.8	46
arena_elo	1508	1439
arena_text_factuality	1486	1434
Artificial Analysis Coding Index	76.5	69.1
browsecomp	88	73.2
coding_arena_elo	1626	1431
critpt	28.6	16.6
DeepSWE	70	54.4
gdpval	61.9	52.9
GDPval-AA (Elo)	1932	1395
GDPval-AA v2	1760	1189
Global-MMLU-Lite	93.3	88.4
GPQA Diamond	92.6	90.8
HLE	64.5	38.6
HLE (with tools)	64.5	45.1
IFBench	63.5	79.2
livebench_agentic_coding	62.2	46.8
livebench_coding	86	75
livebench_data_analysis	80.5	79.3
livebench_instruction_following	75.8	65.5
livebench_language	90.7	79.2
livebench_math	96	86.8
livebench_reasoning	89.7	86.6
MCP Atlas	84.7	69
OmniScience Accuracy	65.3	40.4
OmniScience Non-Hallucination	36.4	15.6
scicode	60.2	49.9
simplebench	81.9	46.3
simpleqa_verified	68.3	45
strongreject	98.7	97.4
SWE-bench Multilingual	86.6	73.3
SWE-bench Pro	80	52.6
SWE-bench Verified	95	79
Tau 3 Banking	26.8	22.9
TauBench V3 - Banking	38.1	39.4
Terminal-Bench 2.1	88	82.7
Terminal-Bench Hard	62.9	38.6
toolathlon	61.7	70.3
τ²-Bench Telecom (AA run)	98.5	95.6
τ³-Bench	26.8	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.