SShogo Bench

Kimi K2 Thinking

Thinking Kimi model for slower research passes, planning, and hard technical questions

Shogo Agent Index

Estimated#70 of 92
30.3range 19.8–40.9 · 23% of weight measured
Agentic coding
32
23% measured
Long-horizon autonomy
37
50% measured
Knowledge-work agents
26
estimated
Computer use
20
estimated
Security & context learning
33
50% measured

Price

Input
$0.63/M
Output
$2.65/M
Cache read
$0.16/M
3:1 blend
$1.14/M
Agent-effective
$0.35/M

Cheapest host for agents: Novita at $0.33/M.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 66
Terminal-Bench 2.035.7%8%Terminus 2
DeepSWEest.6%predicted 25
FrontierSWEest.5%predicted 4
CursorBenchest.4%predicted 27
GSO-Benchest.4%predicted 9
METR Time Horizons59.2%10%source
Vending-Bench 2est.6%predicted 13
PostTrainBenchest.4%predicted 17
APEX-Agentsest.8%predicted 25
GDPvalest.7%predicted 31
Remote Labor Indexest.5%predicted 1
DeepResearch Benchest.5%predicted 45
OSWorld 2.0est.6%predicted 3
OSWorldest.4%predicted 45
Cybenchest.5%predicted 48
CL-bench17.6%5%source

Other benchmarks

BenchmarkCategoryScoreSource
GPQA DiamondReasoning84.2%source
OTIS Mock AIMEMath83.1%source
SimpleBenchReasoning39.6%source
WebDev ArenaCoding1337source

API hosts

Live prices and uptime for every host serving Kimi K2 Thinking via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Novita$0.60$2.50$0.15$0.33100.0%bf16