SShogo Bench

GPT-5.2

Reliable GPT generation for broad coding, writing, and tool-assisted product work

Shogo Agent Index

Supported#44 of 92
46.4range 39.5–53.2 · 58% of weight measured
Agentic coding
49
57% measured
Long-horizon autonomy
59
80% measured
Knowledge-work agents
37
68% measured
Computer use
36
estimated
Security & context learning
46
50% measured

Price

Input
$1.59/M
Output
$12.73/M
Cache read
$0.16/M
3:1 blend
$4.38/M
Agent-effective
$0.96/M

Cheapest host for agents: OpenAI at $0.53/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verified73.8%8%source
Terminal-Bench 2.064.9%8%Droid
DeepSWEest.6%predicted 48
FrontierSWEest.5%predicted 15
CursorBenchest.4%predicted 37
GSO-Bench27.5%4%OpenHands
METR Time Horizons75.3%10%source
Vending-Bench 2$3,5916%source
PostTrainBenchest.4%predicted 25
APEX-Agentsest.8%predicted 44
GDPval49.7%7%source
Remote Labor Index2.5%5%source
DeepResearch Bench41.1%5%source
OSWorld 2.0est.6%predicted 9
OSWorldest.4%predicted 75
Cybenchest.5%predicted 74
CL-bench18.2%5%source

Other benchmarks

BenchmarkCategoryScoreSource
AlgoTuneCoding2.05×source
ARC-AGIReasoning86.2%source
ARC-AGI-2Reasoning52.9%source
Chess PuzzlesReasoning48.0%source
DTBenchKnowledge90.9%source
FrontierMath Tiers 1-3Math67.4%source
FrontierMath Tier 4Math31.7%source
GPQA DiamondReasoning91.4%source
Humanity's Last ExamReasoning27.8%source
LMCAKnowledge43.9%source
MirrorCodeCoding32.2%source
OTIS Mock AIMEMath96.1%source
ProofBenchMath15.0%source
SimpleBenchReasoning45.8%source
SimpleQA VerifiedKnowledge37.1%source
VPCTMultimodal84.0%source
WebDev ArenaCoding1480source
WeirdML v3Coding5.0%codex_cli

API hosts

Live prices and uptime for every host serving GPT-5.2 via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
OpenAI$0.88$7.00$0.087$0.53100.0%—
Azure$1.75$14.00$0.17$1.0699.9%—
OpenAI$1.75$14.00$0.17$1.06100.0%—
OpenAI$3.50$28.00$0.35$2.12100.0%—