Fathom-4B-AIO-GGUF

Fathom-4B-AIO-GGUF

Fathom-DeepResearch is a dual-model agentic system consisting of Fathom-Search-4B and Fathom-Synthesizer-4B, optimized for long-horizon web search, evidence verification, and synthesis into citation-dense reports. These 4B-parameter models are trained to browse, extract, verify, and reason over live web content, setting state-of-the-art open-weight performance on search-heavy tasks (SimpleQA, FRAMES, WebWalkerQA, Seal0) and surpassing closed-source DeepResearch agents like Claude, Grok, Perplexity, and GPT-4o on DeepResearch-Bench. The system leverages a multi-agent self-play pipeline to generate datasets like DuetQA, enforces live web search with multi-hop QA pairs, and employs RAPO (Reward-Aware Policy Optimization) and novel steerable, step-level rewards for stable multi-turn RL with verifiable rewards. The DeepResearch plan-then-write protocol ensures rigorous decomposition, evidence mapping, and insight generation for end-to-end report synthesis. The released stack includes a web-agent search tool server built on Jina-AI, Crawl4AI, Trafilatura, and Serper.dev, supporting high-volume, asynchronous search with URL handlers for YouTube, PDFs, Reddit, and more. The models and code are licensed under MIT to promote open research and community collaboration, with a focus on democratizing complex web-enhanced LLMs for research, development, and AI practitioners worldwide.

Fathom-4B Models GGUF

Model Files

Fathom-Search-4B

Fathom-Synthesizer-4B

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):