SShogo Bench

Gemini 4 Carbon

Google's most capable reasoning model for complex software engineering, deep math and science, high factual accuracy, and long-horizon autonomous agents

Shogo Agent Index

Supported#2 of 92
73.6range 67.2–80 · 27% of weight measured
Agentic coding
82
26% measured
Long-horizon autonomy
80
50% measured
Knowledge-work agents
63
32% measured
Computer use
68
estimated
Security & context learning
63
estimated

Price

Input
$2.00/M
Output
$10.00/M
Cache read
$0.10/M
3:1 blend
$4.00/M
Agent-effective
$0.84/M

Cheapest host for agents: Google AI Studio at $0.42/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 90
Terminal-Bench 2.0est.8%predicted 97
DeepSWEest.6%predicted 84
FrontierSWE61.2%5%proximus · $61.93/task
CursorBench60.4%4%source · $9.22/task
GSO-Benchest.4%predicted 80
METR Time Horizonsest.10%predicted 87
Vending-Bench 2$16,4846%source
PostTrainBench41.0%4%OpenCode
APEX-Agents84.2%8%source
GDPvalest.7%predicted 73
Remote Labor Indexest.5%predicted 24
DeepResearch Benchest.5%predicted 53
OSWorld 2.0est.6%predicted 50
OSWorldest.4%predicted 96
Cybenchest.5%predicted 95
CL-benchest.5%predicted 31

Other benchmarks

BenchmarkCategoryScoreSource
ARC-AGIReasoning98.5%source
ARC-AGI-2Reasoning94.2%source
Chess PuzzlesReasoning78.0%source
CritPtReasoning36.0%source
FrontierMath Tiers 1-3Math90.5%source
FrontierMath Tier 4Math95.1%source
GPQA DiamondReasoning96.5%source
OTIS Mock AIMEMath100.0%source
SciCodeCoding64.2%source
SimpleBenchReasoning92.6%source
SimpleQA VerifiedKnowledge77.7%source

API hosts

Live prices and uptime for every host serving Gemini 4 Carbon via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Google AI Studio$1.00$5.00$0.050$0.42100.0%—
Google$2.00$10.00$0.10$0.8452.6%—
Google AI Studio$2.00$10.00$0.10$0.8499.9%—
Google$3.60$18.00$0.18$1.5258.7%—
Google AI Studio$3.60$18.00$0.18$1.5292.3%—