openpangu/openPangu-Embedded-7B-DeepDiver

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Apr 4, 2026Architecture:Transformer Featherless Exclusive Cold

openPangu-Embedded-7B-DeepDiver is a 7 billion parameter model from openPangu, designed as an Agent for deep information acquisition and processing. It supports native Multi-Agent Systems (MAS) for complex knowledge question answering and long-form report writing, with a context length of 32768 tokens. The model excels at answering multi-step complex questions and generating extensive articles and reports, offering adaptive modes for QA or long-form writing based on user queries.

Loading preview...

openPangu-Embedded-7B-DeepDiver: An Agent for Deep Information Processing

openPangu-Embedded-7B-DeepDiver is a 7 billion parameter model from the openPangu series, specifically engineered as an Agent for advanced information acquisition and processing. It features native Multi-Agent System (MAS) support, enabling it to tackle complex knowledge-based question answering and generate comprehensive long-form reports.

Key Capabilities

  • Complex QA Mode: Capable of answering intricate knowledge-based questions involving over 100 steps.
  • Long-Form Writing Mode: Designed to generate extensive articles and reports exceeding 30,000 characters.
  • Adaptive Mode: Automatically switches between QA and long-form writing based on the user's query.
  • Tool Integration: Supports a wide array of external MCP tools for web search, URL crawling, file operations (read/write), document QA, and content extraction, alongside built-in cognitive tools like think and reflect.

Performance Highlights

Evaluated on specialized benchmarks, DeepDiver demonstrates capabilities in complex question answering:

  • BrowseComp-zh: 18.3 Acc
  • BrowseComp-en: 8.3 Acc
  • XBench-DeepSearch: 39.0 Acc

Further details on long-form writing evaluations are available in the technical report.

Good for

  • Developers building sophisticated AI agents requiring deep information retrieval and synthesis.
  • Applications needing automated generation of detailed research reports or extensive articles.
  • Scenarios demanding multi-step, complex question answering with integrated web search and content processing capabilities.

Finetunes
1 models