apodex/Apodex-1.0-2B-SFT
Apodex/Apodex-1.0-2B-SFT is a 2.3 billion parameter, 32768-token context length model developed by Apodex AI, based on the Qwen3.5 architecture. It is specifically fine-tuned as a verification-centric agent for deep research tasks, designed to operate within an asynchronous agent team for auditable and robust information gathering. This model excels at complex research by preserving general knowledge while enhancing deep-research capabilities through a unique post-training recipe.
Loading preview...
Apodex-1.0-2B-SFT: A Verification-Centric Agent for Deep Research
Apodex-1.0-2B-SFT is a 2.3 billion parameter model from Apodex AI, built upon the Qwen3.5 base and fine-tuned with a 32768-token context length. Its core innovation lies in its verification-centric approach to deep research. While the standalone model functions as a standard tool-using ReAct agent, its full potential is realized in "heavy-duty mode" as part of an asynchronous agent team, Apodex-1.0-H.
Key Capabilities
- Verification-Centric Agent Team: Operates with specialized sub-agents for retrieval and verification, routing reports through a shared evidence pool, and feeding a global verifier to produce highly reliable answers. This system can coordinate up to 150 sub-agents over 15,000 steps.
- Auditable by Construction: Every claim in the final report is backed by an explicit evidence chain and independently audited, ensuring transparency and retractability.
- Preserves General Capabilities: The model's deep-research focus is additive, not a trade-off. It maintains strong performance across general knowledge, mathematics, instruction-following, coding, and long-context tasks, tracking closely with its matched-size Qwen3.5 base.
- State-of-the-Art Performance: Apodex-1.0-H achieves new benchmarks on deep-research suites, including 90.3 on BrowseComp and 94.4 on DeepSearchQA.
Good For
- Complex Deep Research: Ideal for tasks requiring extensive information gathering, cross-referencing, and verification, where accuracy and auditability are paramount.
- Agentic Workflows: Designed for native function calling and integration into agentic systems, leveraging its specialized tool-use and reasoning capabilities.
- Mission-Critical Tasks: Suitable for applications where robust, evidence-backed conclusions are essential, minimizing hallucination through its verification architecture.