apodex/Apodex-1.0-4B-SFT
Apodex/Apodex-1.0-4B-SFT is a 4.5 billion parameter instruction-tuned model developed by Apodex AI, built upon the Qwen3.5 base architecture with a 32768 token context length. It is specifically designed as a verification-centric agent for deep research, excelling in tasks requiring extensive evidence gathering and auditing. The model's post-training recipe focuses on preserving general capabilities while substantially enhancing its deep-research proficiency, making it suitable for complex, auditable intelligence tasks.
Loading preview...
Apodex-1.0-4B-SFT: A Verification-Centric Agent for Deep Research
Apodex-1.0-4B-SFT is a 4.5 billion parameter model from Apodex AI, fine-tuned from the Qwen3.5 base with a 32768 token context length. It is engineered as a verification-centric agent for deep research, designed to operate as a standard tool-using ReAct agent. When deployed in its "heavy-duty mode" (Apodex-1.0-H), it functions as an asynchronous agent team with specialized sub-agents for retrieval and verification, feeding a global verifier that reasons over an assembled evidence graph.
Key Capabilities
- Verification-centric agent team: Utilizes an orchestrator to dispatch a team of specialized sub-agents for parallel exploration and a global verifier to audit assembled evidence, coordinating up to 150 sub-agents over 15,000 steps in a single task.
- Auditable by construction: Every claim in the final report is traceable to an explicit evidence chain and independently checked, with a report pool recording all findings, verdicts, and interventions.
- Preserves general capabilities: Post-training (SFT, agentic DPO, RL) enhances deep-research capabilities without compromising the base model's general knowledge, coding, reasoning, and instruction-following abilities.
- Strong performance on deep-research benchmarks: Apodex-1.0-H achieves state-of-the-art results on benchmarks like BrowseComp (90.3), DeepSearchQA (94.4), and FrontierScience-Research (46.7).
Good For
- Complex deep-research tasks requiring high accuracy and verifiability.
- Applications demanding auditable intelligence with explicit evidence chains.
- Agentic workflows that benefit from parallel exploration and robust verification mechanisms.
- Scenarios where preserving general LLM capabilities alongside specialized research functions is crucial.