Which is better, Claude Fable 5 or GPT-5.5?

Across 63 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 50 and GPT-5.5 leads on 11 — 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-5.5

AnthropicvsOpenAI63 shared benchmarks50–11 head-to-head

Benchmark	Claude Fable 5	GPT-5.5
AA Agentic Index	56.6	47.4
AA Intelligence	62.1	56.3
AA-LCR	76.7	79
AA-Omniscience	43.3	20.5
Agents' Last Exam	25.7	26.6
apex_agents	59.2	38.5
arc_agi_1	90.5	95
ARC-AGI-2	76.8	85
arena_elo	1508	1482
arena_text_factuality	1486	1482
arena_vision	1307	1283
Artificial Analysis Coding Index	76.5	74.9
browsecomp	88	84.4
coding_arena_elo	1626	1457
critpt	28.6	27.1
DeepSWE	70	70
DeepSWE 1.1	70	67
Finance Agent v2	56.3	51.8
GDP (Surge AI)	29.8	16.7
gdpval	61.9	49.5
GDPval-AA (Elo)	1932	1769
GDPval-AA v2	1760	1509
GDPval-AA v2 (Elo)	1747	1491
GPQA Diamond	92.6	93.6
healthbench_professional	66	51.8
HiL-Bench	56.3	39.7
HLE	64.5	52.2
HLE (with tools)	64.5	52.2
IFBench	63.5	75.9
JobBench	57.4	38.3
Legal Agent Benchmark	13.3	2.1
Legal Agent Benchmark (Harvey's Held-Out Set)	13.3	2.1
livebench	78.3	80.7
livebench_agentic_coding	62.2	54
livebench_coding	86	82.2
livebench_data_analysis	80.5	81.6
livebench_instruction_following	75.8	70.7
livebench_language	90.7	87.4
livebench_math	96	95.9
livebench_reasoning	89.7	89.7
MCP Atlas	84.7	82.8
MCP Mark Verified	87.4	92.9
MLS Bench Lite	49.9	35.5
MMMU-Pro	84.2	83.2
officeqa_pro	69.9	60.9
OmniScience Accuracy	65.3	58
OmniScience Non-Hallucination	36.4	12
OSWorld-Verified	85	78.7
PostTrainBench	41.4	28.4
Program Bench	76.8	70.8
scicode	60.2	56.1
simplebench	81.9	69
simpleqa_verified	68.3	63.1
SpreadsheetBench 2	34.7	29.1
SWE-bench Pro	80	58.6
SWE-bench Verified	95	80.6
SWE-Marathon	35	14
TauBench V3 - Banking	38.1	39
Terminal-Bench 2.1	88	84.3
Terminal-Bench Hard	62.9	60.6
toolathlon	61.7	55.6
τ²-Bench Telecom (AA run)	98.5	93.9
τ³-Bench	26.8	31.3

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.