SShogo Bench

Kimi K2.5

Earlier Kimi frontier model for long-context agents, coding, and multimodal work

Shogo Agent Index

Supported#63 of 92
36.5range 28.3–44.8 · 31% of weight measured
Agentic coding
38
46% measured
Long-horizon autonomy
45
30% measured
Knowledge-work agents
30
estimated
Computer use
28
40% measured
Security & context learning
39
50% measured

Price

Input
$0.60/M
Output
$3.00/M
Cache read
$0.10/M
3:1 blend
$1.20/M
Agent-effective
$0.31/M

Cheapest host for agents: Novita at $0.29/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 2.043.2%8%Terminus 2
DeepSWEest.6%predicted 33
FrontierSWEest.5%predicted 7
CursorBenchest.4%predicted 30
GSO-Benchest.4%predicted 14
METR Time Horizonsest.10%predicted 68
Vending-Bench 2$1,1986%source
PostTrainBenchest.4%predicted 20
APEX-Agentsest.8%predicted 31
GDPvalest.7%predicted 36
Remote Labor Indexest.5%predicted 2
DeepResearch Benchest.5%predicted 46
OSWorld 2.0est.6%predicted 4
OSWorld63.3%4%Kimi-K2.5
Cybenchest.5%predicted 58
CL-bench19.3%5%source

Other benchmarks

BenchmarkCategoryScoreSource
ARC-AGIReasoning65.3%source
ARC-AGI-2Reasoning11.8%source
CritPtReasoning3.1%source
Fiction.LiveBenchLong context86.1%source
GPQA DiamondReasoning87.6%source
Humanity's Last ExamReasoning24.4%source
OTIS Mock AIMEMath92.2%source
SciCodeCoding49.0%source
SimpleBenchReasoning46.8%source
SimpleQA VerifiedKnowledge34.3%source
WebDev ArenaCoding1436source

API hosts

Live prices and uptime for every host serving Kimi K2.5 via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
Novita$0.57$2.85$0.095$0.2999.0%—
SiliconFlow$0.50$2.50$0.15$0.3199.7%int4
AtlasCloud$0.49$2.50$0.20$0.3599.6%int4
Amazon Bedrock$0.60$3.00—$0.71100.0%—