Which is better, Claude Fable 5 or Gemini 3.1 Pro?

Across 68 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 54 and Gemini 3.1 Pro leads on 14 — 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 Gemini 3.1 Pro

AnthropicvsGoogle68 shared benchmarks54–14 head-to-head

Benchmark	Claude Fable 5	Gemini 3.1 Pro
AA Agentic Index	56.6	23
AA Intelligence	62.1	47.7
AA-LCR	76.7	79
AA-Omniscience	43.3	32.9
apex_agents	59.2	33.5
arc_agi_1	90.5	98
ARC-AGI-2	76.8	77.1
arena_elo	1508	1486
arena_text_factuality	1486	1471
arena_vision	1307	1280
Artificial Analysis Coding Index	76.5	68.8
baby_vision_with_python	90.5	68.3
Blueprint-Bench 2	38.6	26.5
browsecomp	88	85.9
BrowseComp (w/ Ctx)	88	85.9
charxiv_rq	89.4	89.9
coding_arena_elo	1626	1447
critpt	28.6	17.7
deepsearchqa_f1	94.2	81.9
DeepSWE	70	10
DeepSWE 1.1	70	12
Finance Agent v2	56.3	43
FORTRESS (Adversarial)	96	65.2
FORTRESS (Benign)	55.1	98
gdpval	61.9	23.3
GDPval-AA (Elo)	1932	1317
GDPval-AA v2	1760	962
Global-MMLU-Lite	93.3	92.7
GPQA Diamond	92.6	94.3
HiL-Bench	56.3	35.3
HLE	64.5	51.4
HLE (with tools)	64.5	51.6
IFBench	63.5	77.1
Legal Agent Benchmark	13.3	0
livebench	78.3	79.9
livebench_agentic_coding	62.2	44.1
livebench_coding	86	76.5
livebench_data_analysis	80.5	78.5
livebench_instruction_following	75.8	79.1
livebench_language	90.7	85.4
livebench_math	96	91
livebench_reasoning	89.7	84
mathvision	94.8	89.8
MCP Atlas	84.7	78.2
MMMU-Pro	84.2	83
officeqa_pro	69.9	42.9
OmniScience Accuracy	65.3	55.3
OmniScience Non-Hallucination	36.4	50.1
OSWorld-Verified	85	76.2
PostTrainBench	41.4	21.6
Program Bench	76.8	39.5
scicode	60.2	59
simplebench	81.9	79.6
simpleqa_verified	68.3	77.3
strongreject	98.7	98
SWE-bench Multilingual	86.6	76.9
SWE-bench Pro	80	54.2
SWE-bench Verified	95	80.6
SWE-Marathon	35	4
SWEBench Pro (Public)	80	54.2
Tau 3 Banking	26.8	16.5
TauBench V3 - Banking	38.1	21.4
Terminal Bench 2.1 (Best Harness)	84.6	73.8
Terminal-Bench 2.1	88	74
Terminal-Bench Hard	62.9	68.5
toolathlon	61.7	48.8
τ²-Bench Telecom (AA run)	98.5	99.3
τ³-Bench	26.8	67.1

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.