OpenResearcher/OpenResearcher-30B-A3B

TEXT GENERATIONPricing:Input $0.2 / Output $0.8Concurrent Unit Cost:2Model Size:30BQuant:FP8Context Size:32kPublished:Feb 3, 2026License:mitArchitecture:Transformer0.1K Open Weights Featherless Exclusive Cold

OpenResearcher/OpenResearcher-30B-A3B is a 30 billion parameter agentic large language model developed by OpenResearcher, fine-tuned from NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16. Designed for long-horizon deep research, it was trained on a 96K OpenResearcher dataset with over 100 turns, distilled from GPT-OSS-120B using native browser tools. This model achieves 54.8% accuracy on BrowseComp-Plus, outperforming several larger models in deep research tasks.

Loading preview...

OpenResearcher-30B-A3B: Agentic Deep Research Model

OpenResearcher-30B-A3B is a 30 billion parameter agentic large language model specifically engineered for long-horizon deep research tasks. It is fine-tuned from the NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16 model, leveraging a comprehensive 96K OpenResearcher dataset. This dataset, notable for its 100+ turns, was created by distilling GPT-OSS-120B using native browser tools, enabling the model to handle complex, multi-step research trajectories.

Key Capabilities & Performance

  • Deep Research Proficiency: Designed to excel in intricate research scenarios requiring extended interaction and information synthesis.
  • Benchmark Performance: Achieves an impressive 54.8% accuracy on the BrowseComp-Plus benchmark.
  • Competitive Edge: Surpasses the performance of models like GPT-4.1, Claude-Opus-4, Gemini-2.5-Pro, DeepSeek-R1, and Tongyi-DeepResearch on deep research benchmarks.
  • Extensive Training Data: Benefits from a large-scale, multi-turn dataset specifically curated for research agent capabilities.

When to Use This Model

  • Automated Research Agents: Ideal for developing AI agents that need to perform complex, multi-step research tasks over long horizons.
  • Information Retrieval & Synthesis: Suitable for applications requiring advanced information gathering, analysis, and summarization from web sources.
  • Benchmarking & Evaluation: A strong candidate for evaluating and advancing deep research capabilities in LLMs, particularly on benchmarks like BrowseComp-Plus, BrowseComp, GAIA, and xbench-DeepSearch.