SShogo Bench

Grok 4.20 (Non-Reasoning)

Grok model for agentic tool use, reasoning, coding, and live assistance

Shogo Agent Index

Estimated#49 of 92
43.9range 33.3–54.5 · 19% of weight measured
Agentic coding
45
23% measured
Long-horizon autonomy
59
30% measured
Knowledge-work agents
34
estimated
Computer use
32
estimated
Security & context learning
45
50% measured

Price

Input
$1.25/M
Output
$2.50/M
Cache read
$0.20/M
3:1 blend
$1.56/M
Agent-effective
$0.46/M

Cheapest host for agents: xAI at $0.46/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 75
Terminal-Bench 2.057.3%8%Grok CLI
DeepSWEest.6%predicted 42
FrontierSWEest.5%predicted 11
CursorBenchest.4%predicted 34
GSO-Benchest.4%predicted 21
METR Time Horizonsest.10%predicted 72
Vending-Bench 2$4,6636%source
PostTrainBenchest.4%predicted 23
APEX-Agentsest.8%predicted 39
GDPvalest.7%predicted 42
Remote Labor Indexest.5%predicted 3
DeepResearch Benchest.5%predicted 47
OSWorld 2.0est.6%predicted 7
OSWorldest.4%predicted 69
Cybenchest.5%predicted 68
CL-bench22.2%5%source

Other benchmarks

BenchmarkCategoryScoreSource
ARC-AGIReasoning89.5%source
ARC-AGI-2Reasoning65.1%source
Chess PuzzlesReasoning29.0%source
DTBenchKnowledge90.1%source
FrontierMath Tiers 1-3Math44.9%source
FrontierMath Tier 4Math17.1%source
GPQA DiamondReasoning89.3%source
LMCAKnowledge38.7%source
OTIS Mock AIMEMath92.2%source
ProofBenchMath14.0%source
SimpleQA VerifiedKnowledge30.2%source
WebDev ArenaCoding1374source

API hosts

Live prices and uptime for every host serving Grok 4.20 (Non-Reasoning) via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
xAI$1.25$2.50$0.20$0.46100.0%—
xAI$1.25$2.50$0.20$0.46100.0%—
xAI$2.50$5.00$0.40$0.92100.0%—
xAI$2.50$5.00$0.40$0.92——