Which is better, Claude Fable 5 or Claude Opus 4.6?

Across 52 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 45 and Claude Opus 4.6 leads on 7 — 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 Claude Opus 4.6

AnthropicvsAnthropic52 shared benchmarks45–7 head-to-head

Benchmark	Claude Fable 5	Claude Opus 4.6
AA Agentic Index	56.6	67.6
AA Intelligence	62.1	44.9
AA-LCR	76.7	74.3
AA-Omniscience	43.3	13.7
arc_agi_1	90.5	93
ARC-AGI-2	76.8	68.8
arena_elo	1508	1497
arena_text_factuality	1486	1487
arena_vision	1307	1300
Artificial Analysis Coding Index	76.5	48.1
baby_vision_with_python	90.5	38.4
browsecomp	88	86.8
charxiv_rq	89.4	69.1
coding_arena_elo	1626	1545
critpt	28.6	12.6
deepsearchqa_f1	94.2	91.3
gdpval	61.9	55.9
GDPval-AA (Elo)	1932	1619
GPQA Diamond	92.6	91.3
HiL-Bench	56.3	38.3
HLE	64.5	53.1
HLE (with tools)	64.5	62.7
IFBench	63.5	62.5
Legal Agent Benchmark	13.3	4.2
livebench	78.3	76.3
livebench_agentic_coding	62.2	49
livebench_coding	86	78.2
livebench_data_analysis	80.5	69.9
livebench_instruction_following	75.8	63.3
livebench_language	90.7	83.3
livebench_math	96	89.3
livebench_reasoning	89.7	88.7
mathvision	94.8	71.2
MCP Atlas	84.7	76.8
MMMU-Pro	84.2	77.3
officeqa_pro	69.9	57.1
OmniScience Accuracy	65.3	47
OmniScience Non-Hallucination	36.4	37.2
OSWorld-Verified	85	72.7
scicode	60.2	52
simplebench	81.9	67.6
simpleqa_verified	68.3	46.5
swe_bench_multimodal	54.1	27.1
SWE-bench Multilingual	86.6	77.8
SWE-bench Pro	80	57.3
SWE-bench Verified	95	80.8
Terminal-Bench Hard	62.9	65.4
toolathlon	61.7	56.8
Toolathlon Avg turns	19.8	16.9
Toolathlon Pass@∞	55.6	47.2
τ²-Bench Telecom (AA run)	98.5	99.3
τ³-Bench	26.8	72.4

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.