SShogo Bench

Qwen2.5-72B

Qwen2.5-72B from Alibaba (Qwen).

Shogo Agent Index

Estimated#91 of 92
12.1range 8.5–15.7 · 14% of weight measured
Agentic coding
11
estimated
Long-horizon autonomy
20
50% measured
Knowledge-work agents
12
estimated
Computer use
2
40% measured
Security & context learning
9
estimated

Price

No public API price.

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

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 38
Terminal-Bench 2.0est.8%predicted 2
DeepSWEest.6%predicted 4
FrontierSWEest.5%predicted 0
CursorBenchest.4%predicted 12
GSO-Benchest.4%predicted 1
METR Time Horizons35.8%10%source
Vending-Bench 2est.6%predicted 0
PostTrainBenchest.4%predicted 8
APEX-Agentsest.8%predicted 5
GDPvalest.7%predicted 10
Remote Labor Indexest.5%predicted 0
DeepResearch Benchest.5%predicted 38
OSWorld 2.0est.6%predicted 0
OSWorld5.0%4%qwen2.5-vl-72b-instruct (100 steps)
Cybenchest.5%predicted 9
CL-benchest.5%predicted 10

Other benchmarks

BenchmarkCategoryScoreSource
BALROGLong-horizon16.2%source
DTBenchKnowledge62.9%source
GeoBenchMultimodal62.0%source
GPQA DiamondReasoning49.1%source
LMCAKnowledge13.4%source
MATH Level 5Math63.2%source
MMLULegacy85.3%source
OTIS Mock AIMEMath8.1%source
TheAgentCompanyKnowledge work5.7%source