SShogo Bench

Claude Fable 5.2

Claude model for demanding reasoning and long-horizon agentic work

Shogo Agent Index

Supported#1 of 92
76.4range 71.7–81.2 · 32% of weight measured
Agentic coding
86
26% measured
Long-horizon autonomy
78
50% measured
Knowledge-work agents
67
52% measured
Computer use
77
estimated
Security & context learning
65
estimated

Price

Input
$8.00/M
Output
$40.00/M
Cache read
$0.20/M
3:1 blend
$16.00/M
Agent-effective
$3.21/M

Cheapest host for agents: Anthropic at $3.21/M.

Agent-effective = 20 input tokens per output token, 85% cache hits. Why.

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 92
Terminal-Bench 2.0est.8%predicted 98
DeepSWEest.6%predicted 88
FrontierSWE69.9%5%proximus · $168.78/task
CursorBench60.0%4%source · $32.79/task
GSO-Benchest.4%predicted 88
METR Time Horizonsest.10%predicted 90
Vending-Bench 2$9,6286%source
PostTrainBench44.6%4%Claude Code
APEX-Agents80.1%8%source
GDPvalest.7%predicted 78
Remote Labor Index41.3%5%source
DeepResearch Benchest.5%predicted 55
OSWorld 2.0est.6%predicted 63
OSWorldest.4%predicted 98
Cybenchest.5%predicted 97
CL-benchest.5%predicted 34

Other benchmarks

BenchmarkCategoryScoreSource
ARC-AGIReasoning98.5%source
ARC-AGI-2Reasoning93.8%source
Chess PuzzlesReasoning79.0%source
CritPtReasoning35.7%source
DTBenchKnowledge99.2%source
FrontierMath Tiers 1-3Math95.1%source
FrontierMath Tier 4Math100.0%source
GPQA DiamondReasoning96.0%source
LMCAKnowledge70.6%source
OTIS Mock AIMEMath99.7%source
ProofBenchMath100.0%source
SciCodeCoding67.7%source
SimpleBenchReasoning94.9%source
SimpleQA VerifiedKnowledge94.0%source
WeirdML v3Coding43.1%claude_code

API hosts

Live prices and uptime for every host serving Claude Fable 5.2 via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Anthropic$8.00$40.00$0.20$3.21100.0%—
Azure$8.00$40.00$0.20$3.21100.0%—
Google$8.00$40.00$0.20$3.21100.0%—
Anthropic$16.00$80.00$0.40$6.42100.0%—