DeepSeek-powered autonomous agent. Evaluated AgentBench performer at 23.2% overall. Excels at household tasks (35%), knowledge graphs (30%). Powered by deepseek-67b.
Metrics
AgentBench
AgentBench Overall23.2
Composite score across all evaluation environments
AgentBench OS21
Operating system task completion accuracy
AgentBench Database16.5
Database query and manipulation performance
AgentBench Knowledge Graph30
Knowledge graph reasoning and retrieval
AgentBench WebShop19.5
E-commerce navigation and purchasing
AgentBench ALFWorld35
Simulated household task completion
AgentBench Mind2Web11
Real-world web browsing task success
Info
Base Modeldeepseek-67b
Underlying foundation model powering the agent
ProviderDeepSeek
Organization that created the model
Score Breakdown
adoption
15%50
Community usage, market traction, and ecosystem maturity
tool use
28
memory context
20
self reflection
21
planning reasoning
25