Moonshot's Kimi-Researcher Launches for Deep Research Tasks
Moonshot's Kimi-Researcher Launches for Deep Research Tasks
Moonshot Dark Side has officially unveiled Kimi-Researcher, its first AI-powered deep research agent, now in limited internal testing. The new model leverages end-to-end autonomous reinforcement learning (agentic RL) to provide users with efficient, in-depth research capabilities.
Advanced Autonomous Research Capabilities
When tackling complex queries, Kimi-Researcher demonstrates remarkable autonomy:
- Performs 23 steps of reasoning per task on average
- Plans 74 search keywords per inquiry
- Evaluates 206 URLs, retaining only the top 3.2% of highest-quality content
The system goes beyond simple information retrieval by:
- Automatically calling tools like browsers and code interpreters
- Processing raw data into actionable insights
- Generating comprehensive reports with traceable sources
Benchmark Performance and Real-World Applications
To validate its capabilities, developers subjected Kimi-Researcher to the rigorous Humanity's Last Exam (HLE) benchmark, which spans hundreds of professional domains including:
- Mathematics and physics
- Medical research
- Political science and history The model achieved impressive scores of 26.9% Pass@1 and 40.17% Pass@4 accuracy, outperforming several established AI systems.
In practical scenarios, Kimi-Researcher has proven valuable for:
- Algorithm engineers seeking high-value benchmarks
- Business analysts researching industry trends
- Legal professionals comparing international data privacy laws The system can produce 10,000+ word reports with approximately 26 quality references, complete with shareable interactive visualizations.
Technical Innovation and Availability
The model's unique architecture features:
- Zero-structure design: No complex prompts or preset workflows
- Self-adaptation: Learns entirely through trial-and-error reinforcement learning This approach enables superior performance when handling conflicting information or adapting to environmental changes.
The service is currently in limited beta testing. Interested users can apply for access at kimi.com and activate the "Deep Research" feature after approval.
Key Points:
- Moonshot Dark Side launches AI research agent Kimi-Researcher in beta testing
- System autonomously plans searches, filters content, and generates detailed reports
- Achieved top-tier performance on challenging Humanity's Last Exam benchmark
- Currently available through limited access program at kimi.com
