LiteResearcher-4B by simplex-ai-inc is a 4 billion parameter deep research agent built on the Qwen3ForCausalLM architecture, featuring a 262,144 token context window. Trained with scalable agentic reinforcement learning, it excels at complex research tasks by iteratively thinking, searching the web, visiting pages for evidence, and synthesizing answers. This model matches Claude-4.5-Sonnet on GAIA and outperforms open-source models up to 8x larger, making it highly efficient for web-based information retrieval and synthesis.
Loading preview...
LiteResearcher-4B: A Compact Deep Research Agent
LiteResearcher-4B is a 4 billion parameter deep research agent developed by simplex-ai-inc, designed to perform complex information retrieval and synthesis tasks. Built upon the Qwen3ForCausalLM architecture (specifically Qwen3-4B-Thinking base), it leverages a substantial 262,144 token maximum context length.
Key Capabilities & Differentiators
- Agentic Reinforcement Learning: Trained via a two-stage difficulty-aware curriculum RL in a virtual world environment, enabling scalable agentic behavior.
- ReAct-style Operation: Functions as a ReAct agent, using
<think>,<tool_call>, and<answer>tags to structure its reasoning process. It iteratively thinks, searches the web via Google, visits webpages to extract evidence, and then formulates answers. - Exceptional Performance for Size: Despite its compact 4B parameters, LiteResearcher-4B achieves performance comparable to or exceeding much larger models. It matches Claude-4.5-Sonnet on the GAIA-Text benchmark (71.3%) and outperforms open-source models up to 8 times its size on benchmarks like Xbench-DS (78.0%) and WebWalkerQA (72.7%).
- Extensive Training Data: Utilizes a co-constructed training corpus of over 32 million webpages and 1 million domains, covering five atomic search capabilities.
Ideal Use Cases
LiteResearcher-4B is particularly well-suited for applications requiring:
- Automated Web Research: Efficiently gathering and synthesizing information from the internet.
- Complex Question Answering: Answering intricate questions that require multi-step reasoning and external tool use.
- Resource-Constrained Environments: Deploying powerful research capabilities where larger models are impractical due to computational or memory limitations.