Which is better, Claude Fable 5 or GPT-5.4?

Across 50 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 42 and GPT-5.4 leads on 8 — 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 GPT-5.4

AnthropicvsOpenAI50 shared benchmarks42–8 head-to-head

Benchmark	Claude Fable 5	GPT-5.4
AA Agentic Index	56.6	58.2
AA Intelligence	62.1	53.1
AA-LCR	76.7	77.7
AA-Omniscience	43.3	5.8
arc_agi_1	90.5	93.7
ARC-AGI-2	76.8	74
arena_elo	1508	1477
arena_text_factuality	1486	1476
arena_vision	1307	1277
Artificial Analysis Coding Index	76.5	71.1
baby_vision_with_python	90.5	80.2
browsecomp	88	82.7
charxiv_rq	89.4	82.8
coding_arena_elo	1626	1457
critpt	28.6	23.4
deepsearchqa_f1	94.2	78.6
DeepSWE 1.1	70	52
gdpval	61.9	50.1
GDPval-AA (Elo)	1932	1674
GPQA Diamond	92.6	93
HiL-Bench	56.3	9.7
HLE	64.5	43.7
HLE (with tools)	64.5	52.1
IFBench	63.5	73.9
Legal Agent Benchmark	13.3	0.4
livebench	78.3	80.3
livebench_agentic_coding	62.2	53.8
livebench_coding	86	77.5
livebench_data_analysis	80.5	79.3
livebench_instruction_following	75.8	70.2
livebench_language	90.7	82.6
livebench_math	96	94.2
livebench_reasoning	89.7	88.1
mathvision	94.8	92
MCP Atlas	84.7	70.6
MMMU-Pro	84.2	81.2
officeqa_pro	69.9	51.1
OmniScience Accuracy	65.3	50.9
OmniScience Non-Hallucination	36.4	17.4
OSWorld-Verified	85	75
scicode	60.2	56.6
simpleqa_verified	68.3	45.3
SWE-bench Multilingual	86.6	71.7
SWE-bench Pro	80	59.1
TauBench V3 - Banking	38.1	39.6
Terminal-Bench 2.1	88	78.3
Terminal-Bench Hard	62.9	57.6
toolathlon	61.7	54.6
τ²-Bench Telecom (AA run)	98.5	87.1
τ³-Bench	26.8	72.9

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.