SShogo Bench

Inkling 2 Small

Multimodal MoE reasoning model (594B total, 17B active) for agentic coding, vision, and knowledge work

Shogo Agent Index

Estimated#24 of 92
55.2range 42.3–68.2 · 13% of weight measured
Agentic coding
62
14% measured
Long-horizon autonomy
66
estimated
Knowledge-work agents
42
32% measured
Computer use
44
estimated
Security & context learning
54
estimated

Price

Input
$0.30/M
Output
$1.00/M
Cache read
$0.060/M
3:1 blend
$0.48/M
Agent-effective
$0.14/M

Cheapest host for agents: PrimeIntellect at $0.070/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 82
Terminal-Bench 2.0est.8%predicted 84
DeepSWEest.6%predicted 61
FrontierSWE25.1%5%proximus · $11.84/task
CursorBenchest.4%predicted 43
GSO-Benchest.4%predicted 45
METR Time Horizonsest.10%predicted 80
Vending-Bench 2est.6%predicted 68
PostTrainBenchest.4%predicted 29
APEX-Agents46.8%8%source
GDPvalest.7%predicted 55
Remote Labor Indexest.5%predicted 7
DeepResearch Benchest.5%predicted 50
OSWorld 2.0est.6%predicted 17
OSWorldest.4%predicted 86
Cybenchest.5%predicted 84
CL-benchest.5%predicted 24

Other benchmarks

BenchmarkCategoryScoreSource
ARC-AGIReasoning96.5%source
ARC-AGI-2Reasoning81.7%source
SciCodeCoding52.3%source

API hosts

Live prices and uptime for every host serving Inkling 2 Small via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
PrimeIntellect$0.15$0.50$0.030$0.07099.9%fp4
Cloudflare$0.30$1.00$0.030$0.11100.0%—
DeepInfra$0.28$0.95$0.029$0.1199.4%fp4
Modal$0.30$1.00$0.030$0.11100.0%nvfp4
Together$0.30$1.00$0.030$0.11100.0%—
Nebius$0.30$1.00$0.060$0.1497.3%fp4